IBM 3090 Model 600J compared with modern desktop PC's.

[Sternbach, William]

Hi,

I hope its ok if I share some interesting research I did on the capacity
and specs of IBM 3090 Model 600J computer
which I had been impressed by in the early 1990's as being able to
handle 700 TSO online users doing Cobol compiles,
multiple 24/7 CICS regions, huge numbers of concurrently running Cobol
batch programs doing massive I/O during
nightly batch cycles, etc.

I came to some surprising conclusions after doing this research,
specifically at how slow and low capacity
this high end mainframe 3090 600J computer really was compared with
today's desktop PC's.

Please see: http://ukcc.uky.edu/~ukccinfo/ibm3090.html (for the specs of
model installed in Kentucky).

Please see: http://freespace.virgin.net/roy.longbottom/mips.htm (which
lists the 600J's CPU speed as only 69 Mhz each).

I hope you will forgive my use of Mhz and Ram (as I know they're not
directly comparible).

IBM 3090 600J had: 6 CPU's running at 69 Mhz each, 200 GB total hard
disk space, 1.5 GB Ram.

My Desktop PC has: Intel Core 2 Duo 2,666 Mhz, 1,000 GB total hard disk
space, 4 GB Ram.

Then, I researched a long lived Fortran benchmark that actually ran on a
wide range of computers (see below):
Mdbnch - Molecular Chemistry benchmark:
http://www.fisica.uniud.it/~ercolessi/Mdbnch/info.html
and the results are interesting. To summarize, my desktop PC ran this
benchmark
in less than 1 second, while the IBM 3090 600J ran this benchmark in 76
seconds.
The Cray computer ran it in: 15 seconds, and some old Digital Vaxes
took: 8,000 seconds.

MACHINE, COMPILER, COMPILATION OPTIONS TIME
DATE
--------------------------------------------------------------- ------
-------
Intel Dual Core E6700 2.66 Ghz Gnu g77 3.2.3, -Wall -O3 -s
0.937500 02Jun07
AMD Athlon 64 bit 4000+ 2.6GHZ Gnu g77 3.2.3, -Wall -O3 -s
1.390625 18Jul06
Intel Pentium IV 2.8 GHZ Gnu g77 3.2.3, -Wall -O3 -s 2.3125s
09Dec05
Intel Pentium IV 2.4 GHZ (533MHZ bus/memory), Gnu g77 3.2, -O3 2.47s
19Mar03
Intel Pentium IV 630 3.0 GHZ Gnu g77 3.2.3, -Wall -O3 -s
2.515625 21Jul06
Intel Pentium III 1066 MHZ Gnu g77 3.2.3, -Wall -O3 -s
5.047258 10Dec05
Intel Pentium III 733 MHZ (133MHZ bus/memory), Gnu g77 3.2, -O3 7.13s
17Feb03
Intel Pentium III 600 MHZ (100MHZ bus/memory), Gnu g77 3.2, -O3 8.03s
12Feb03
Intel PentiumPro 200MHz/256K, Win95, Intel F77 2.4, -G6 -Qxi ... 22.6 s
22Oct97
Intel-MMX 266MHz, OS/2 V4 FP#10, Watcom F77^32 11.a [^] ... 30.9 s
25May99
Intel P133/60ns, Linux, glibc 2, g77 0.5.21-1997081 [^].... 59.6 s
21Aug97
Intel P133/60ns, Linux, glibc 2, f2c+gcc 2.8.0beta [^] .... 62.9 s
21Aug97
Intel Pentium 100MHz, OS/2+DOS, Watcom F. 9.5, /FPI87 /OX . 90.7 s
08Feb96
Intel Pentium/90MHz, Linux, G77 0.5.15 + GCC i2.6.3 (*) .. 125 s
20Jul95
Intel Pentium/90MHz, Linux, G77 0.5.15 + GCC i2.6.3 (+) .. 134 s
20Jul95
Intel Pentium/90MHz, Microway Fortran NDP 486, mf486 -on . 132 s
22Dec94
Intel i860 40 MHz (board Microway on a 386 PC) ............... 172 s
13Feb91
Intel 486DX4/100MHz, Win95, g77 0.5.19+gcc 2.7.2.1.f.1 [o] 210 s
18Jun98
Intel 486DX3/100MHz, Microway Fortran NDP 486, mf486 -on . 232 s
22Dec94
Intel 486DX2/66MHz, Microway Fortran NDP 486, mf486 -on .. 358 s
22Dec94
Micron Pentium Pro 200MHz, Linux, g77 2.7.2.f.1, -O6 .......... 51.6 s
18Jul96

IBM 3090-J, VM/SP, VS Fortran 2.4.0, OPT(3), vector ........... 76 s
07Jun90
IBM 3090-J, VM/SP, VS Fortran 2.4.0, OPT(3), scalar ........... 92 s
07Jun90
IBM 3090-E, MVS/XA, VS Fortran 2.3.0, OPT(3), vector ......... 99 s
20Dec89
IBM 3090-E, MVS/XA, VS Fortran 2.3.0, OPT(3), scalar ......... 118 s
20Dec89
IBM 3081, VM/CMS, VS Fortran 2.3.0, OPT(3) ................... 360 s
20Dec89
IBM 3081, VM/CMS, VS Fortran 2.3.0, OPT(2) ................... 361 s
20Dec89
IBM 3081, VM/CMS, VS Fortran 2.3.0, OPT(1) ................... 523 s
20Dec89

VAX 9000-410, Vax Fortran, scalar ............................ 101 s
18Apr91
VAX 4000-100, Vax Fortran V5.7-133 /OPT ...................... 148 s
07Apr93
VAX 4000-100, Vax Fortran-HPO V1.3-163 /OPT .................. 156 s
07Apr93
VAX 4000-100, Vax Fortran V5.7-133 /NOOPT .................... 190 s
07Apr93
VAX 4000-100, Vax Fortran-HPO V1.3-163 /NOOPT ................ 190 s
07Apr93
VAX 6000-510, Vax Fortran, scalar ............................ 398 s
Vax 6410, VMS, Vax Fortran 5.4-79, opt ....................... 648 s
06Feb91
Vax 8800, VMS, Vax Fortran 5.4, opt .......................... 854 s
14Mar90
Vax 6310, VMS, Vax Fortran 5.4, opt ......................... 1326 s
13Mar90
VaxStation 3100/38, VMS, Vax Fortran 5.4, opt ............... 1378 s
13Mar90
Vax 6210, VMS, Vax Fortran 4.7, opt ......................... 1873 s
VaxStation 3500, VMS, Vax Fortran 4.7, opt................... 2017 s
VaxStation 3500, Ultrix, Berkeley f77 1.0, opt .............. 2275 s
13Mar90
Microvax II, VMS, Vax Fortran 5.4, opt ...................... 6941 s
14Mar90
Vax 750, VMS, Vax Fortran 4.7, opt ......................... 6973 s
VaxStation 2000, VMS, Vax Fortran 5.4, opt .................. 7255 s
14Mar90
VaxStation 2000, Ultrix, f77, opt ........................... 8306 s
23Dec88

Cray T90, Unicos 9.0.1., cf77 6.2.3.0 ......................... 15.6 s
17Jan98
Cray T90, Unicos 9.0, f90 V.3.0, -dp -O 1 ..................... 22.9 s
04Mar98
Cray C90, Unicos 7.C.3, cf77 6.0.1.4, -Wf"-a stack" ........... 28.4 s
17Dec93
Cray C90, Unicos 7.C.3, cf77 6.0.1.4, -Zv -Wf"-a static" ...... 31.5 s
17Dec93
Cray C90, Unicos 7.C.3, cf77 6.0.1.4, -Zv -Wf"-a stack" ....... 31.6 s
17Dec93
Cray C90, Unicos 7.C.3, cf77 6.0.1.4, -Wf"-a stack -o novector" 40.9 s
17Dec93
Cray YMP8/432, cf77 5.0.4.4,-Zv -Wf"-a stack -o agg,bl,la,inl". 36.5 s
24Jan93
Cray YMP8/432, cf77 5.0.4.4,-Zv -Wf"-a stack" ................. 37.4 s
24Jan93
Cray YMP8/432, cf77 5.0.4.4,-Zv -Wf"-a static" ................ 37.9 s
24Jan93
Cray YMP8/432, cf77 5.0.4.4,-Wf"-a stack" ..................... 43.5 s
24Jan93
Cray YMP8/432, cf77 5.0.4.4,-Wf"-a stack -o novector" ......... 63.9 s
24Jan93
Cray XMP/48, cft77 2.0, 46-bit integers, vector ............... 55 s
21Dec88
Cray XMP/48, cft77 2.0, 46-bit integers, scalar ............... 73 s
21Dec88
Cray-2 4 processors, 256 Mw, cf77 5.0, vector ................. 83 s
23Oct91
Cray-2 4 processors, 256 Mw, cf77 5.0, scalar ................ 116 s
23Oct91

Sun Ultra 10 (333MHz), Solaris 2.6, f77 SC4.2 [o] ............. 9.90s
25May99
[o] -fast -xO5 -xtarget=native -xarch=v8plus
Sun Ultra 2 (300MHz), Solaris 2.6, f77 SC4.2 [o] .............. 10.1 s+
16Jul97
[o] -fast -xO5 -xtarget=native -xarch=v8plus -stackvar -xdepend
Sun Ultra 30 (300 MHz), Solaris 2.6, Sun f77 4.2 [o] .......... 10.5 s
16Jun98
[o] -fast -xO5 -xtarget=3Dultra2 -xarch=3Dv8plus -fns

- Bill

[Sternbach, William]

i,

Sorry to post this twice, I realigined the chart below so it would
display better...

Bill

I hope its ok if I share some interesting research I did on the capacity
and specs of IBM 3090 Model 600J computer which I had been impressed by
in the early 1990's as being able to handle 700 TSO online users doing
Cobol compiles, multiple 24/7 CICS regions, huge numbers of concurrently
running Cobol batch programs doing massive I/O during nightly batch
cycles, etc.

I came to some surprising conclusions after doing this research,
specifically
at how slow and low capacity this high end mainframe 3090 600J computer
really
was compared with today's desktop PC's.

Please see: http://ukcc.uky.edu/~ukccinfo/ibm3090.html
(for the specs of model installed in Kentucky).

Please see: http://freespace.virgin.net/roy.longbottom/mips.htm
(which lists the 600J's CPU speed as only 69 Mhz each).

I hope you will forgive my use of Mhz and Ram
(as I know they're not directly comparible).

IBM 3090 600J had:
6 CPU's running at 69 Mhz each, 200 GB total hard disk space, 1.5
GB Ram.

My Desktop PC has:
Intel Core 2 Duo 2,666 Mhz, 1,000 GB total hard disk space, 4 GB
Ram.

Then, I researched a long lived Fortran benchmark that actually ran on a
wide
range of computers (see below): Mdbnch - Molecular Chemistry benchmark:
http://www.fisica.uniud.it/~ercolessi/Mdbnch/info.html
and the results are interesting. To summarize, my desktop PC ran this
benchmark
in less than 1 second, while the IBM 3090 600J ran this benchmark in 76
seconds.
The Cray computer ran it in: 15 seconds, and some old Digital Vaxes
took: 8,000 seconds.

MACHINE, COMPILER, COMPILATION OPTIONS TIME DATE
--------------------------------------- ------- ------ -------
Intel Dual Core E6700 2.66 Ghz 0.937500 02Jun07
AMD Athlon 64 bit 4000+ 2.6GHZ 1.390625 18Jul06
Intel Pentium IV 2.8 GHZ 2.3125s 09Dec05
Intel Pentium IV 2.4 GHZ 2.47s 19Mar03
Intel Pentium IV 630 3.0 GHZ 2.515625 21Jul06
Intel Pentium III 1066 MHZ 5.047258 10Dec05
Intel Pentium III 733 MHZ 7.13s 17Feb03
Intel Pentium III 600 MHZ 8.03s 12Feb03
Intel PentiumPro 200MHz/256K 22.6 s 22Oct97
Intel-MMX 266MHz 30.9 s 25May99
Intel P133/60ns 59.6 s 21Aug97
Intel P133/60ns 62.9 s 21Aug97
Intel Pentium 100MHz 90.7 s 08Feb96
Intel Pentium/90MHz 125 s 20Jul95
Intel Pentium/90MHz 134 s 20Jul95
Intel Pentium/90MHz 132 s 22Dec94
Intel i860 40 MHz 172 s 13Feb91
Intel 486DX4/100MHz 210 s 18Jun98
Intel 486DX3/100MHz 232 s 22Dec94
Intel 486DX2/66MHz 358 s 22Dec94
Micron Pentium Pro 200MHz 51.6 s 18Jul96

IBM 3090-J, VM/SP, OPT(3), vector 76 s 07Jun90
IBM 3090-J, VM/SP, OPT(3), scalar 92 s 07Jun90
IBM 3090-E, MVS/XA,OPT(3), vector 99 s 20Dec89
IBM 3090-E, MVS/XA,OPT(3), scalar 118 s 20Dec89
IBM 3081, VM/CMS, OPT(3) 360 s 20Dec89
IBM 3081, VM/CMS, OPT(2) 361 s 20Dec89
IBM 3081, VM/CMS, OPT(1) 523 s 20Dec89

VAX 9000-410, Vax Fortran, scalar 101 s 18Apr91
VAX 4000-100, Vax Fortran V5.7-133 148 s 07Apr93
VAX 4000-100, Vax Fortran-HPO 156 s 07Apr93
VAX 4000-100, Vax Fortran V5.7-133 190 s 07Apr93
VAX 4000-100, Vax Fortran-HPO 190 s 07Apr93
VAX 6000-510, Vax Fortran, scalar 398 s
Vax 6410, VMS, Vax Fortran 5.4-79 648 s 06Feb91
Vax 8800, VMS, Vax Fortran 5.4 854 s 14Mar90
Vax 6310, VMS, Vax Fortran 5.4 1326 s 13Mar90
VaxStation 3100/38, VMS, 1378 s 13Mar90
Vax 6210, VMS, Vax Fortran 4.7, opt 1873 s
VaxStation 3500, VMS, Vax Fortran 2017 s
VaxStation 3500, Ultrix, Berkeley 2275 s 13Mar90
Microvax II, VMS, Vax Fortran 5.4 6941 s 14Mar90
Vax 750, VMS, Vax Fortran 4.7 6973 s
VaxStation 2000, VMS 7255 s 14Mar90
VaxStation 2000, Ultrix, f77, opt 8306 s 23Dec88

Cray T90, Unicos 9.0.1., cf77 15.6 s 17Jan98
Cray T90, Unicos 9.0, 22.9 s 04Mar98
Cray C90, Unicos 7.C.3 28.4 s 17Dec93
Cray C90, Unicos 7.C.3 31.5 s 17Dec93
Cray C90, Unicos 7.C.3 31.6 s 17Dec93
Cray C90, Unicos 7.C.3 40.9 s 17Dec93
Cray YMP8/432, cf77 36.5 s 24Jan93
Cray YMP8/432, cf77 37.4 s 24Jan93
Cray YMP8/432, cf77 37.9 s 24Jan93
Cray YMP8/432, cf77 43.5 s 24Jan93
Cray YMP8/432, cf77 63.9 s 24Jan93
Cray XMP/48, cft77 2.0 55 s 21Dec88
Cray XMP/48, cft77 2.0 73 s 21Dec88
Cray-2 4 processors, 256 Mw 83 s 23Oct91
Cray-2 4 processors, 256 Mw 116 s 23Oct91

Sun Ultra 10 (333MHz), Solaris 2.6, 9.90s 25May99
Sun Ultra 2 (300MHz), Solaris 2.6, 10.1 s+ 16Jul97
Sun Ultra 30 (300 MHz), Solaris 2.6 10.5 s 16Jun98

- Bill

[peter_flass]

On a CPU basis only, the mainframes lose. Where they shine is in
keeping lots of I/O going all the time. I don't believe PC hardware is
there yet. This is why IBM doesn't usually quote MIPS. I'd still
like to see a PeeCee handle the workload quoted.

Post by Sternbach, William
Hi,
I hope its ok if I share some interesting research I did on the capacity
and specs of IBM 3090 Model 600J computer
which I had been impressed by in the early 1990's as being able to
handle 700 TSO online users doing Cobol compiles,
multiple 24/7 CICS regions, huge numbers of concurrently running Cobol
batch programs doing massive I/O during
nightly batch cycles, etc.
I came to some surprising conclusions after doing this research,
specifically at how slow and low capacity
this high end mainframe 3090 600J computer really was compared with
today's desktop PC's.
Please see: http://ukcc.uky.edu/~ukccinfo/ibm3090.html (for the specs of
model installed in Kentucky).
Please see: http://freespace.virgin.net/roy.longbottom/mips.htm (which
lists the 600J's CPU speed as only 69 Mhz each).
I hope you will forgive my use of Mhz and Ram (as I know they're not
directly comparible).
IBM 3090 600J had: 6 CPU's running at 69 Mhz each, 200 GB total hard
disk space, 1.5 GB Ram.
My Desktop PC has: Intel Core 2 Duo 2,666 Mhz, 1,000 GB total hard disk
space, 4 GB Ram.
Then, I researched a long lived Fortran benchmark that actually ran on a
http://www.fisica.uniud.it/~ercolessi/Mdbnch/info.html
and the results are interesting. To summarize, my desktop PC ran this
benchmark
in less than 1 second, while the IBM 3090 600J ran this benchmark in 76
seconds.
The Cray computer ran it in: 15 seconds, and some old Digital Vaxes
took: 8,000 seconds.
MACHINE, COMPILER, COMPILATION OPTIONS TIME
DATE
--------------------------------------------------------------- ------
-------
Intel Dual Core E6700 2.66 Ghz Gnu g77 3.2.3, -Wall -O3 -s
0.937500 02Jun07
AMD Athlon 64 bit 4000+ 2.6GHZ Gnu g77 3.2.3, -Wall -O3 -s
1.390625 18Jul06
Intel Pentium IV 2.8 GHZ Gnu g77 3.2.3, -Wall -O3 -s 2.3125s
09Dec05
Intel Pentium IV 2.4 GHZ (533MHZ bus/memory), Gnu g77 3.2, -O3 2.47s
19Mar03
Intel Pentium IV 630 3.0 GHZ Gnu g77 3.2.3, -Wall -O3 -s
2.515625 21Jul06
Intel Pentium III 1066 MHZ Gnu g77 3.2.3, -Wall -O3 -s
5.047258 10Dec05
Intel Pentium III 733 MHZ (133MHZ bus/memory), Gnu g77 3.2, -O3 7.13s
17Feb03
Intel Pentium III 600 MHZ (100MHZ bus/memory), Gnu g77 3.2, -O3 8.03s
12Feb03
Intel PentiumPro 200MHz/256K, Win95, Intel F77 2.4, -G6 -Qxi ... 22.6 s
22Oct97
Intel-MMX 266MHz, OS/2 V4 FP#10, Watcom F77^32 11.a [^] ... 30.9 s
25May99
Intel P133/60ns, Linux, glibc 2, g77 0.5.21-1997081 [^].... 59.6 s
21Aug97
Intel P133/60ns, Linux, glibc 2, f2c+gcc 2.8.0beta [^] .... 62.9 s
21Aug97
Intel Pentium 100MHz, OS/2+DOS, Watcom F. 9.5, /FPI87 /OX . 90.7 s
08Feb96
Intel Pentium/90MHz, Linux, G77 0.5.15 + GCC i2.6.3 (*) .. 125 s
20Jul95
Intel Pentium/90MHz, Linux, G77 0.5.15 + GCC i2.6.3 (+) .. 134 s
20Jul95
Intel Pentium/90MHz, Microway Fortran NDP 486, mf486 -on . 132 s
22Dec94
Intel i860 40 MHz (board Microway on a 386 PC) ............... 172 s
13Feb91
Intel 486DX4/100MHz, Win95, g77 0.5.19+gcc 2.7.2.1.f.1 [o] 210 s
18Jun98
Intel 486DX3/100MHz, Microway Fortran NDP 486, mf486 -on . 232 s
22Dec94
Intel 486DX2/66MHz, Microway Fortran NDP 486, mf486 -on .. 358 s
22Dec94
Micron Pentium Pro 200MHz, Linux, g77 2.7.2.f.1, -O6 .......... 51.6 s
18Jul96
IBM 3090-J, VM/SP, VS Fortran 2.4.0, OPT(3), vector ........... 76 s
07Jun90
IBM 3090-J, VM/SP, VS Fortran 2.4.0, OPT(3), scalar ........... 92 s
07Jun90
IBM 3090-E, MVS/XA, VS Fortran 2.3.0, OPT(3), vector ......... 99 s
20Dec89
IBM 3090-E, MVS/XA, VS Fortran 2.3.0, OPT(3), scalar ......... 118 s
20Dec89
IBM 3081, VM/CMS, VS Fortran 2.3.0, OPT(3) ................... 360 s
20Dec89
IBM 3081, VM/CMS, VS Fortran 2.3.0, OPT(2) ................... 361 s
20Dec89
IBM 3081, VM/CMS, VS Fortran 2.3.0, OPT(1) ................... 523 s
20Dec89
VAX 9000-410, Vax Fortran, scalar ............................ 101 s
18Apr91
VAX 4000-100, Vax Fortran V5.7-133 /OPT ...................... 148 s
07Apr93
VAX 4000-100, Vax Fortran-HPO V1.3-163 /OPT .................. 156 s
07Apr93
VAX 4000-100, Vax Fortran V5.7-133 /NOOPT .................... 190 s
07Apr93
VAX 4000-100, Vax Fortran-HPO V1.3-163 /NOOPT ................ 190 s
07Apr93
VAX 6000-510, Vax Fortran, scalar ............................ 398 s
Vax 6410, VMS, Vax Fortran 5.4-79, opt ....................... 648 s
06Feb91
Vax 8800, VMS, Vax Fortran 5.4, opt .......................... 854 s
14Mar90
Vax 6310, VMS, Vax Fortran 5.4, opt ......................... 1326 s
13Mar90
VaxStation 3100/38, VMS, Vax Fortran 5.4, opt ............... 1378 s
13Mar90
Vax 6210, VMS, Vax Fortran 4.7, opt ......................... 1873 s
VaxStation 3500, VMS, Vax Fortran 4.7, opt................... 2017 s
VaxStation 3500, Ultrix, Berkeley f77 1.0, opt .............. 2275 s
13Mar90
Microvax II, VMS, Vax Fortran 5.4, opt ...................... 6941 s
14Mar90
Vax 750, VMS, Vax Fortran 4.7, opt ......................... 6973 s
VaxStation 2000, VMS, Vax Fortran 5.4, opt .................. 7255 s
14Mar90
VaxStation 2000, Ultrix, f77, opt ........................... 8306 s
23Dec88
Cray T90, Unicos 9.0.1., cf77 6.2.3.0 ......................... 15.6 s
17Jan98
Cray T90, Unicos 9.0, f90 V.3.0, -dp -O 1 ..................... 22.9 s
04Mar98
Cray C90, Unicos 7.C.3, cf77 6.0.1.4, -Wf"-a stack" ........... 28.4 s
17Dec93
Cray C90, Unicos 7.C.3, cf77 6.0.1.4, -Zv -Wf"-a static" ...... 31.5 s
17Dec93
Cray C90, Unicos 7.C.3, cf77 6.0.1.4, -Zv -Wf"-a stack" ....... 31.6 s
17Dec93
Cray C90, Unicos 7.C.3, cf77 6.0.1.4, -Wf"-a stack -o novector" 40.9 s
17Dec93
Cray YMP8/432, cf77 5.0.4.4,-Zv -Wf"-a stack -o agg,bl,la,inl". 36.5 s
24Jan93
Cray YMP8/432, cf77 5.0.4.4,-Zv -Wf"-a stack" ................. 37.4 s
24Jan93
Cray YMP8/432, cf77 5.0.4.4,-Zv -Wf"-a static" ................ 37.9 s
24Jan93
Cray YMP8/432, cf77 5.0.4.4,-Wf"-a stack" ..................... 43.5 s
24Jan93
Cray YMP8/432, cf77 5.0.4.4,-Wf"-a stack -o novector" ......... 63.9 s
24Jan93
Cray XMP/48, cft77 2.0, 46-bit integers, vector ............... 55 s
21Dec88
Cray XMP/48, cft77 2.0, 46-bit integers, scalar ............... 73 s
21Dec88
Cray-2 4 processors, 256 Mw, cf77 5.0, vector ................. 83 s
23Oct91
Cray-2 4 processors, 256 Mw, cf77 5.0, scalar ................ 116 s
23Oct91
Sun Ultra 10 (333MHz), Solaris 2.6, f77 SC4.2 [o] ............. 9.90s
25May99
[o] -fast -xO5 -xtarget=native -xarch=v8plus
Sun Ultra 2 (300MHz), Solaris 2.6, f77 SC4.2 [o] .............. 10.1 s+
16Jul97
[o] -fast -xO5 -xtarget=native -xarch=v8plus -stackvar -xdepend
Sun Ultra 30 (300 MHz), Solaris 2.6, Sun f77 4.2 [o] .......... 10.5 s
16Jun98
[o] -fast -xO5 -xtarget=3Dultra2 -xarch=3Dv8plus -fns
- Bill

[Dave Wade]

Desktop PC with local IDE disk might not keep up. Decent server with fibre
SAN with dedicated disk and no redundancy will blow 3090 i/o out of the
water. Lets face it the 3390Mod27 disk is usually just the same SAN disk
that is used on PC servers. If you don't believe me down load some of the
Exchange benchmarks from the Microsoft web site, or Notes benchmarks from
the IBM web site and look at the i/o they do. I know it's a "benchmark" but
several manufacturers have reported simulating serving over 30,000 mail
users on a PC whilst maintaining sub-second response times. Work out the I/O
required and compare to 3090.



Mind you, you do need to do things properly. Just because its possible, as
shown by the many mainframe sites that put their DASD on SAN, it does not
mean its done properly for PC servers. All too often the I/O system is
compromised by tuning for capacity over performance.



Mind you, getting a "balanced system" is important. I personally think one
reason the DEC Alpha chip achieved so little penetration of the PC server
market place was that at the time it was "too fast" for the I/O in use.
Moving to the Alpha from Pentium achieved very little because the disk
subsystems of the time couldn't keep up.



Dave



-----Original Message-----
From: hercules-390-***@public.gmane.org [mailto:hercules-390-***@public.gmane.org] On
Behalf Of peter_flass
Sent: 15 June 2007 23:26
To: hercules-390-***@public.gmane.org
Subject: [hercules-390] Re: IBM 3090 Model 600J compared with modern desktop
PC's.



On a CPU basis only, the mainframes lose. Where they shine is in
keeping lots of I/O going all the time. I don't believe PC hardware is
there yet. This is why IBM doesn't usually quote MIPS. I'd still
like to see a PeeCee handle the workload quoted.

--- In hercules-390@ <mailto:hercules-390%40yahoogroups.com>
yahoogroups.com, "Sternbach, William "

Post by Sternbach, William
Hi,
I hope its ok if I share some interesting research I did on the capacity
and specs of IBM 3090 Model 600J computer
which I had been impressed by in the early 1990's as being able to
handle 700 TSO online users doing Cobol compiles,
multiple 24/7 CICS regions, huge numbers of concurrently running Cobol
batch programs doing massive I/O during
nightly batch cycles, etc.
I came to some surprising conclusions after doing this research,
specifically at how slow and low capacity
this high end mainframe 3090 600J computer really was compared with
today's desktop PC's.
Please see: http://ukcc. <http://ukcc.uky.edu/~ukccinfo/ibm3090.html>

uky.edu/~ukccinfo/ibm3090.html (for the specs of

Post by Sternbach, William
model installed in Kentucky).
Please see: http://freespace.

<http://freespace.virgin.net/roy.longbottom/mips.htm>
virgin.net/roy.longbottom/mips.htm (which

Post by Sternbach, William
lists the 600J's CPU speed as only 69 Mhz each).
I hope you will forgive my use of Mhz and Ram (as I know they're not
directly comparible).
IBM 3090 600J had: 6 CPU's running at 69 Mhz each, 200 GB total hard
disk space, 1.5 GB Ram.
My Desktop PC has: Intel Core 2 Duo 2,666 Mhz, 1,000 GB total hard disk
space, 4 GB Ram.
Then, I researched a long lived Fortran benchmark that actually ran on a
http://www.fisica.

<http://www.fisica.uniud.it/~ercolessi/Mdbnch/info.html>
uniud.it/~ercolessi/Mdbnch/info.html

Post by Sternbach, William
and the results are interesting. To summarize, my desktop PC ran this
benchmark
in less than 1 second, while the IBM 3090 600J ran this benchmark in 76
seconds.
The Cray computer ran it in: 15 seconds, and some old Digital Vaxes
took: 8,000 seconds.
MACHINE, COMPILER, COMPILATION OPTIONS TIME
DATE
---------------------------------------------------------- ------
-------
Intel Dual Core E6700 2.66 Ghz Gnu g77 3.2.3, -Wall -O3 -s
0.937500 02Jun07
AMD Athlon 64 bit 4000+ 2.6GHZ Gnu g77 3.2.3, -Wall -O3 -s
1.390625 18Jul06
Intel Pentium IV 2.8 GHZ Gnu g77 3.2.3, -Wall -O3 -s 2.3125s
09Dec05
Intel Pentium IV 2.4 GHZ (533MHZ bus/memory), Gnu g77 3.2, -O3 2.47s
19Mar03
Intel Pentium IV 630 3.0 GHZ Gnu g77 3.2.3, -Wall -O3 -s
2.515625 21Jul06
Intel Pentium III 1066 MHZ Gnu g77 3.2.3, -Wall -O3 -s
5.047258 10Dec05
Intel Pentium III 733 MHZ (133MHZ bus/memory), Gnu g77 3.2, -O3 7.13s
17Feb03
Intel Pentium III 600 MHZ (100MHZ bus/memory), Gnu g77 3.2, -O3 8.03s
12Feb03
Intel PentiumPro 200MHz/256K, Win95, Intel F77 2.4, -G6 -Qxi ... 22.6 s
22Oct97
Intel-MMX 266MHz, OS/2 V4 FP#10, Watcom F77^32 11.a [^] ... 30.9 s
25May99
Intel P133/60ns, Linux, glibc 2, g77 0.5.21-1997081 [^].... 59.6 s
21Aug97
Intel P133/60ns, Linux, glibc 2, f2c+gcc 2.8.0beta [^] .... 62.9 s
21Aug97
Intel Pentium 100MHz, OS/2+DOS, Watcom F. 9.5, /FPI87 /OX . 90.7 s
08Feb96
Intel Pentium/90MHz, Linux, G77 0.5.15 + GCC i2.6.3 (*) .. 125 s
20Jul95
Intel Pentium/90MHz, Linux, G77 0.5.15 + GCC i2.6.3 (+) .. 134 s
20Jul95
Intel Pentium/90MHz, Microway Fortran NDP 486, mf486 -on . 132 s
22Dec94
Intel i860 40 MHz (board Microway on a 386 PC) ............... 172 s
13Feb91
Intel 486DX4/100MHz, Win95, g77 0.5.19+gcc 2.7.2.1.f.1 [o] 210 s
18Jun98
Intel 486DX3/100MHz, Microway Fortran NDP 486, mf486 -on . 232 s
22Dec94
Intel 486DX2/66MHz, Microway Fortran NDP 486, mf486 -on .. 358 s
22Dec94
Micron Pentium Pro 200MHz, Linux, g77 2.7.2.f.1, -O6 .......... 51.6 s
18Jul96
IBM 3090-J, VM/SP, VS Fortran 2.4.0, OPT(3), vector ........... 76 s
07Jun90
IBM 3090-J, VM/SP, VS Fortran 2.4.0, OPT(3), scalar ........... 92 s
07Jun90
IBM 3090-E, MVS/XA, VS Fortran 2.3.0, OPT(3), vector ......... 99 s
20Dec89
IBM 3090-E, MVS/XA, VS Fortran 2.3.0, OPT(3), scalar ......... 118 s
20Dec89
IBM 3081, VM/CMS, VS Fortran 2.3.0, OPT(3) ................... 360 s
20Dec89
IBM 3081, VM/CMS, VS Fortran 2.3.0, OPT(2) ................... 361 s
20Dec89
IBM 3081, VM/CMS, VS Fortran 2.3.0, OPT(1) ................... 523 s
20Dec89
VAX 9000-410, Vax Fortran, scalar ............................ 101 s
18Apr91
VAX 4000-100, Vax Fortran V5.7-133 /OPT ...................... 148 s
07Apr93
VAX 4000-100, Vax Fortran-HPO V1.3-163 /OPT .................. 156 s
07Apr93
VAX 4000-100, Vax Fortran V5.7-133 /NOOPT .................... 190 s
07Apr93
VAX 4000-100, Vax Fortran-HPO V1.3-163 /NOOPT ................ 190 s
07Apr93
VAX 6000-510, Vax Fortran, scalar ............................ 398 s
Vax 6410, VMS, Vax Fortran 5.4-79, opt ....................... 648 s
06Feb91
Vax 8800, VMS, Vax Fortran 5.4, opt .......................... 854 s
14Mar90
Vax 6310, VMS, Vax Fortran 5.4, opt ......................... 1326 s
13Mar90
VaxStation 3100/38, VMS, Vax Fortran 5.4, opt ............... 1378 s
13Mar90
Vax 6210, VMS, Vax Fortran 4.7, opt ......................... 1873 s
VaxStation 3500, VMS, Vax Fortran 4.7, opt................... 2017 s
VaxStation 3500, Ultrix, Berkeley f77 1.0, opt .............. 2275 s
13Mar90
Microvax II, VMS, Vax Fortran 5.4, opt ...................... 6941 s
14Mar90
Vax 750, VMS, Vax Fortran 4.7, opt ......................... 6973 s
VaxStation 2000, VMS, Vax Fortran 5.4, opt .................. 7255 s
14Mar90
VaxStation 2000, Ultrix, f77, opt ........................... 8306 s
23Dec88
Cray T90, Unicos 9.0.1., cf77 6.2.3.0 ......................... 15.6 s
17Jan98
Cray T90, Unicos 9.0, f90 V.3.0, -dp -O 1 ..................... 22.9 s
04Mar98
Cray C90, Unicos 7.C.3, cf77 6.0.1.4, -Wf"-a stack" ........... 28.4 s
17Dec93
Cray C90, Unicos 7.C.3, cf77 6.0.1.4, -Zv -Wf"-a static" ...... 31.5 s
17Dec93
Cray C90, Unicos 7.C.3, cf77 6.0.1.4, -Zv -Wf"-a stack" ....... 31.6 s
17Dec93
Cray C90, Unicos 7.C.3, cf77 6.0.1.4, -Wf"-a stack -o novector" 40.9 s
17Dec93
Cray YMP8/432, cf77 5.0.4.4,-Zv -Wf"-a stack -o agg,bl,la,inl". 36.5 s
24Jan93
Cray YMP8/432, cf77 5.0.4.4,-Zv -Wf"-a stack" ................. 37.4 s
24Jan93
Cray YMP8/432, cf77 5.0.4.4,-Zv -Wf"-a static" ................ 37.9 s
24Jan93
Cray YMP8/432, cf77 5.0.4.4,-Wf"-a stack" ..................... 43.5 s
24Jan93
Cray YMP8/432, cf77 5.0.4.4,-Wf"-a stack -o novector" ......... 63.9 s
24Jan93
Cray XMP/48, cft77 2.0, 46-bit integers, vector ............... 55 s
21Dec88
Cray XMP/48, cft77 2.0, 46-bit integers, scalar ............... 73 s
21Dec88
Cray-2 4 processors, 256 Mw, cf77 5.0, vector ................. 83 s
23Oct91
Cray-2 4 processors, 256 Mw, cf77 5.0, scalar ................ 116 s
23Oct91
Sun Ultra 10 (333MHz), Solaris 2.6, f77 SC4.2 [o] ............. 9.90s
25May99
[o] -fast -xO5 -xtarget=native -xarch=v8plus
Sun Ultra 2 (300MHz), Solaris 2.6, f77 SC4.2 [o] .............. 10.1 s+
16Jul97
[o] -fast -xO5 -xtarget=native -xarch=v8plus -stackvar -xdepend
Sun Ultra 30 (300 MHz), Solaris 2.6, Sun f77 4.2 [o] .......... 10.5 s
16Jun98
[o] -fast -xO5 -xtarget=3Dultra2 -xarch=3Dv8plus -fns
- Bill

[Non-text portions of this message have been removed]

[wgs77]

Hi,

I hope its ok if I make 1 more comment about the IBM 3090 Model 600J.
I was "in Awe" of the limitless resources the IBM 3090 Model 600J
seemed to have - with its 700+ TSO users in ISPF running Cobol
compiles and executing test programs all day, while simultaneously
handling numerous CICS partitions on a 24/7 basis with users all over
the world using CICS which behind the scenes was executing online
Cobol programs doing validations and then reading / updating VSAM,
IMS, or DB2 updates. The users hitting enter would get sub-second
response time.

The discovery I made yesterday was that the IBM 3090 Model 600J's
CPU's only ran at 69 Mhz, and it only had 1.5GB Memory (Ram) and 200
GB total Disk space. My desktop PC: Intel Core 2 Duo E6700, with
2,666 Mhz 64 bit CPU, 4 GB Ram and 1000 GB total disk space makes me
think that my desktop is faster. Then I discovered that the Molecular
Dynamics Mdbnch Fortran benchmark program, which took 76 seconds to
run on an IBM 3090 model 600J, only took less than 1 second to run on
my Intel Core 2 Duo E6700 PC. Since the measurements were in CPU
time, not clock time, I conclude that my Core 2 Duo desktop PC is
faster. I remember the tour I took of the computer room, with the
enormous white boxes representing the CPU's of the 3090 Model 600J and
also the 3380 and 3390 Disk Packs, it had to believe that my desktop
PC is superior. This is a numbing discovery for me, and its taking
some time for it to sink in...

Bill

[Dave Morton]

Your PC is superior to what???
The benchmarks you listed measured only CPU time for a CPU intensive
FORTRAN program if I understood it correctly - and not elapsed
time??? Doesn't elapsed time matter to a user?

Are you saying that your 2007-PC has more power and capability than
the 3090 you worked with 15 years ago?

I'm curious how many TSO users are typically logged on to your PC,
how many CICS users, how many batch jobs running, how many IMS and
DB2 databases are in use, how many printers printing and at whhat
speed, what kind of (RACF) security you have on your PC, memory
protection, tape drives, monitoring software, etc, etc.

My 1997 car can probably outrun a 1930s steam locomotive pulling 50
freight cars cross country, or pulling 10 passenger cars from New
York to LA, with meals served en route. But that doesn't make my
car "more powerful". If it did, I could rent it out to Amtrak even
today, and they could get rid of some of those pesky diesel engines,
freight cars and passenger cars, and save a TON of money!!

Dave M

Post by wgs77
Hi,
I hope its ok if I make 1 more comment about the IBM 3090 Model 600J.
I was "in Awe" of the limitless resources the IBM 3090 Model 600J
seemed to have - with its 700+ TSO users in ISPF running Cobol
compiles and executing test programs all day, while simultaneously
handling numerous CICS partitions on a 24/7 basis with users all over
the world using CICS which behind the scenes was executing online
Cobol programs doing validations and then reading / updating VSAM,
IMS, or DB2 updates. The users hitting enter would get sub-second
response time.
The discovery I made yesterday was that the IBM 3090 Model 600J's
CPU's only ran at 69 Mhz, and it only had 1.5GB Memory (Ram) and 200
GB total Disk space. My desktop PC: Intel Core 2 Duo E6700, with
2,666 Mhz 64 bit CPU, 4 GB Ram and 1000 GB total disk space makes me
think that my desktop is faster. Then I discovered that the

Molecular

Post by wgs77
Dynamics Mdbnch Fortran benchmark program, which took 76 seconds to
run on an IBM 3090 model 600J, only took less than 1 second to run on
my Intel Core 2 Duo E6700 PC. Since the measurements were in CPU
time, not clock time, I conclude that my Core 2 Duo desktop PC is
faster. I remember the tour I took of the computer room, with the
enormous white boxes representing the CPU's of the 3090 Model 600J and
also the 3380 and 3390 Disk Packs, it had to believe that my

desktop

Post by wgs77
PC is superior. This is a numbing discovery for me, and its taking
some time for it to sink in...
Bill

[Gerhard Postpischil]

Post by Dave Morton
My 1997 car can probably outrun a 1930s steam locomotive pulling 50
freight cars cross country, or pulling 10 passenger cars from New
York to LA, with meals served en route. But that doesn't make my
car "more powerful". If it did, I could rent it out to Amtrak even
today, and they could get rid of some of those pesky diesel engines,
freight cars and passenger cars, and save a TON of money!!

1930s steam locomotives routinely ran 100 mph on scheduled
passenger service, and at least one made it to 150 mph on a test
by the New York Central system. Your car may or may not be able
to match that. A fast freight train, running perishables,
usually was limited to 60-70 mph; otherwise average speeds of
15-45 mph were common.

As to pulling power, your car at best might be able to pull 2
passenger cars, provided some weight were added over the pulling
axle. Budd Rail Diesel Cars (RDC) came with 120 HP engines, and
were not able to pull a trailing car in normal service. Many
railroads upgraded to 200-250 HP engines, and pulled one
additional passenger car.

Gerhard Postpischil
Bradford, VT

new e-mail address: gerhardp (at) charter (dot) net

[Dave Morton]

Thanks, Gerhard. I learned something (as usual)!

But of course, having a car pull railroad cars would be cheating...
That would mix "train" and "car" instead of keeping them separate.

For additional mainframe advantages, see the document $OSTL18 in the
FILES section, in the part that addresses "PC People New to
Mainframes" (I know you're obviously not new to mainframes - it's an
interesting read for ANY level, in my opinion).

I'm out of the technology and IBM mainframe loop so I can't update
the document with more current examples, but the existing examples
make the points: Compare apples to apples, compare current to
current, compare additional features of the mainframe, etc.

Comparing a current PC to an old mainframe is fun and interesting (I
enjoyed doing it to the limited extent I could), but even that
excercise requires a healthy dose of reality nearly impossible to
simulate!! :)

Dave M

Post by Gerhard Postpischil

Post by Dave Morton
My 1997 car can probably outrun a 1930s steam locomotive pulling 50
freight cars cross country, or pulling 10 passenger cars from New
York to LA, with meals served en route. But that doesn't make my
car "more powerful". If it did, I could rent it out to Amtrak even
today, and they could get rid of some of those pesky diesel

engines,

Post by Gerhard Postpischil

Post by Dave Morton
freight cars and passenger cars, and save a TON of money!!

1930s steam locomotives routinely ran 100 mph on scheduled
passenger service, and at least one made it to 150 mph on a test
by the New York Central system. Your car may or may not be able
to match that. A fast freight train, running perishables,
usually was limited to 60-70 mph; otherwise average speeds of
15-45 mph were common.
As to pulling power, your car at best might be able to pull 2
passenger cars, provided some weight were added over the pulling
axle. Budd Rail Diesel Cars (RDC) came with 120 HP engines, and
were not able to pull a trailing car in normal service. Many
railroads upgraded to 200-250 HP engines, and pulled one
additional passenger car.
Gerhard Postpischil
Bradford, VT
new e-mail address: gerhardp (at) charter (dot) net

[Dave Jones]

This is getting way off topic, but since trains where brought up, I
thought I'd add this little tidbit....

The folks at the railroads that make up freight train configurations
(i.e., how many freight car and of what types), allocate 1/4 horsepower
per ton of freight to be move. A modern diesel-electric locomotive, like
the ones made today by GE, generate about 4,000 horsepower....so if you
want to know the total weight of a freight train as it passes you at a
crossing, count the number of engines and multiple by 16,000....that
will be an approximation of the total weight of the train, in tons.
Example a four engine train should way around 64,000 tons (4 x 16,000).

Post by Gerhard Postpischil

Post by Dave Morton
My 1997 car can probably outrun a 1930s steam locomotive pulling 50
freight cars cross country, or pulling 10 passenger cars from New
York to LA, with meals served en route. But that doesn't make my
car "more powerful". If it did, I could rent it out to Amtrak even
today, and they could get rid of some of those pesky diesel engines,
freight cars and passenger cars, and save a TON of money!!

1930s steam locomotives routinely ran 100 mph on scheduled
passenger service, and at least one made it to 150 mph on a test
by the New York Central system. Your car may or may not be able
to match that. A fast freight train, running perishables,
usually was limited to 60-70 mph; otherwise average speeds of
15-45 mph were common.
As to pulling power, your car at best might be able to pull 2
passenger cars, provided some weight were added over the pulling
axle. Budd Rail Diesel Cars (RDC) came with 120 HP engines, and
were not able to pull a trailing car in normal service. Many
railroads upgraded to 200-250 HP engines, and pulled one
additional passenger car.
Gerhard Postpischil
Bradford, VT
new e-mail address: gerhardp (at) charter (dot) net

[Dave Jones]

My keyboard seems to have a bad case of dyslexia today....
make that "many freight cars", "to be moved" and "should weight"....

Post by Dave Jones
This is getting way off topic, but since trains where brought up, I
thought I'd add this little tidbit....
The folks at the railroads that make up freight train configurations
(i.e., how many freight car and of what types), allocate 1/4 horsepower
per ton of freight to be move. A modern diesel-electric locomotive, like
the ones made today by GE, generate about 4,000 horsepower....so if you
want to know the total weight of a freight train as it passes you at a
crossing, count the number of engines and multiple by 16,000....that
will be an approximation of the total weight of the train, in tons.
Example a four engine train should way around 64,000 tons (4 x 16,000).

[Rob van der Heij]

Post by wgs77
This is a numbing discovery for me, and its taking
some time for it to sink in...

Efficiency was a key issue with slow machines. Developers don't worry
about that anymore.
I ran 100 Linux virtual machines with apache web server on z/VM on a P/390.
That's rated 3-4 MIPS or so, compare it with a 12 MHz Intel CPU...

Rob

[Mike Schwab]

Where I work, we switched 2 of 8 Shark ESS F20 boxes from Mainframe
ESCON use to RS/6000 AIX FCP. The 4 mainframes kept the controller
boxes 75% busy during backups after midnight for an hour, 30-40% busy
during daytime hours, and about 20% busy evenings during batch
processing. The 14 RS/6000s got the control units up to 8-9% busy
during backups just after midnight, otherwise it was 0, 1, or on
occasion 2% busy during the day.

Post by Dave Wade
Desktop PC with local IDE disk might not keep up. Decent server with fibre
SAN with dedicated disk and no redundancy will blow 3090 i/o out of the
water. Lets face it the 3390Mod27 disk is usually just the same SAN disk
that is used on PC servers. If you don't believe me down load some of the
Exchange benchmarks from the Microsoft web site, or Notes benchmarks from
the IBM web site and look at the i/o they do. I know it's a "benchmark" but
several manufacturers have reported simulating serving over 30,000 mail
users on a PC whilst maintaining sub-second response times. Work out the I/O
required and compare to 3090.
Mind you, you do need to do things properly. Just because its possible, as
shown by the many mainframe sites that put their DASD on SAN, it does not
mean its done properly for PC servers. All too often the I/O system is
compromised by tuning for capacity over performance.
Mind you, getting a "balanced system" is important. I personally think one
reason the DEC Alpha chip achieved so little penetration of the PC server
market place was that at the time it was "too fast" for the I/O in use.
Moving to the Alpha from Pentium achieved very little because the disk
subsystems of the time couldn't keep up.
Dave

[Dave Wade]

Mike,



Sounds typical to me. I wonder what the CPU utilization is like on the AIX
boxes. I bet that's also a lot lower than on the Mainframe. In fact it's
common to use far more server boxes than you really need, to get functional
separation and redundancy, which means folks have a lot of underused
servers, which are ripe for virtualization...

.. so just because the AIX boxes don't hammer the SAN doesn't mean they
couldn't given appropriate software such as Hercules.



I think it's real pity we can never get to test this in real life. Whilst I
don't expect Hercules to be any where near a modern box, I am certain that
given appropriate disk subsystems it would certainly equal the system
performance of all but the fastest 3090.



I also realize that building GCC is not a real test, but does involve a fair
amount of I/O. When I run it on VM/370 on a dual core P4, with fairly old
120gb PATA drives, I can't get the I/O queue length much above 1, but the
"real" CPU core emulating sits solid on 100% with the virtual CPU at around
27 MIPS. It uses QSAM I/O so that's not really what I would expect, and
wonder if any one can say why..



Dave.



P.S. Do you think we are going to see Windows Server Core
(http://redmondmag.com/features/article.asp?EditorialsID=640) running on
BOCHS on mainframe..





-----Original Message-----
From: hercules-390-***@public.gmane.org [mailto:hercules-390-***@public.gmane.org] On
Behalf Of Mike Schwab
Sent: 16 June 2007 05:36
To: hercules-390-***@public.gmane.org
Subject: Re: [hercules-390] Re: IBM 3090 Model 600J compared with modern
desktop PC's.



Where I work, we switched 2 of 8 Shark ESS F20 boxes from Mainframe
ESCON use to RS/6000 AIX FCP. The 4 mainframes kept the controller
boxes 75% busy during backups after midnight for an hour, 30-40% busy
during daytime hours, and about 20% busy evenings during batch
processing. The 14 RS/6000s got the control units up to 8-9% busy
during backups just after midnight, otherwise it was 0, 1, or on
occasion 2% busy during the day.

Post by Dave Wade
Desktop PC with local IDE disk might not keep up. Decent server with fibre
SAN with dedicated disk and no redundancy will blow 3090 i/o out of the
water. Lets face it the 3390Mod27 disk is usually just the same SAN disk
that is used on PC servers. If you don't believe me down load some of the
Exchange benchmarks from the Microsoft web site, or Notes benchmarks from
the IBM web site and look at the i/o they do. I know it's a "benchmark" but
several manufacturers have reported simulating serving over 30,000 mail
users on a PC whilst maintaining sub-second response times. Work out the I/O
required and compare to 3090.
Mind you, you do need to do things properly. Just because its possible, as
shown by the many mainframe sites that put their DASD on SAN, it does not
mean its done properly for PC servers. All too often the I/O system is
compromised by tuning for capacity over performance.
Mind you, getting a "balanced system" is important. I personally think one
reason the DEC Alpha chip achieved so little penetration of the PC server
market place was that at the time it was "too fast" for the I/O in use.
Moving to the Alpha from Pentium achieved very little because the disk
subsystems of the time couldn't keep up.
Dave

--
Mike A Schwab, Springfield IL USA http://geocities.
<http://geocities.com/maschwab/> com/maschwab/ for
software links





[Non-text portions of this message have been removed]

[Paul Raulerson]

Oh, I wouldn't say that. AIX or Linux can and will put a load on a San,
but to be perfectly honest, despite the great specifications on the PCI and
PCI-X busses and their kin, I have not yet seen a non-mainframe size pSeries
or iSeries box able to engage the same kind of load a mainframe can. Note,
the very top of end of both p and I series machines are as close to being a
mainframe as makes no difference, at least in terms of I/O. They are very
different machines from the rack mounted servers most people think of for
pSeries, and the server size boxes most people think of for iSeries
machines.



Mainframes, in particular mainframes running z/OS are elegant. No fooling,
they do things very differently from the typical low end machines. Hercules,
running on a PC, proves that it is not all in the hardware though, as even
the legal OSs, like MVS, can put "a hurting" on the PC's CPU and I/O
capacity. I would love to see it running z/OS, or z/VM, both of which will
run, but are forbidden of course.



z/Linux running under Hercules running under Linux does a respectable job
too.



It isn't in the design parameters for Hercules, but I would love to see
Hercules load as the hypervisor on a PC, without the overhead of a full
Linux load. I am convinced that would produce some amazing results. But
without a modern day legal O/S, it is a futile dream to pursue.



-Paul



From: hercules-390-***@public.gmane.org [mailto:hercules-390-***@public.gmane.org] On
Behalf Of Dave Wade
Sent: Saturday, June 16, 2007 4:12 AM
To: hercules-390-***@public.gmane.org
Subject: RE: [hercules-390] Now somewhat OT Re: IBM 3090 Model 600J compared
with modern desktop PC's.



Mike,

Sounds typical to me. I wonder what the CPU utilization is like on the AIX
boxes. I bet that's also a lot lower than on the Mainframe. In fact it's
common to use far more server boxes than you really need, to get functional
separation and redundancy, which means folks have a lot of underused
servers, which are ripe for virtualization...

.. so just because the AIX boxes don't hammer the SAN doesn't mean they
couldn't given appropriate software such as Hercules.

I think it's real pity we can never get to test this in real life. Whilst I
don't expect Hercules to be any where near a modern box, I am certain that
given appropriate disk subsystems it would certainly equal the system
performance of all but the fastest 3090.

I also realize that building GCC is not a real test, but does involve a fair
amount of I/O. When I run it on VM/370 on a dual core P4, with fairly old
120gb PATA drives, I can't get the I/O queue length much above 1, but the
"real" CPU core emulating sits solid on 100% with the virtual CPU at around
27 MIPS. It uses QSAM I/O so that's not really what I would expect, and
wonder if any one can say why..

Dave.

P.S. Do you think we are going to see Windows Server Core
(http://redmondmag.com/features/article.asp?EditorialsID=640) running on
BOCHS on mainframe..

-----Original Message-----
From: hercules-390-***@public.gmane.org <mailto:hercules-390%40yahoogroups.com>
[mailto:hercules-390-***@public.gmane.org <mailto:hercules-390%40yahoogroups.com>
] On
Behalf Of Mike Schwab
Sent: 16 June 2007 05:36
To: hercules-390-***@public.gmane.org <mailto:hercules-390%40yahoogroups.com>
Subject: Re: [hercules-390] Re: IBM 3090 Model 600J compared with modern
desktop PC's.

Where I work, we switched 2 of 8 Shark ESS F20 boxes from Mainframe
ESCON use to RS/6000 AIX FCP. The 4 mainframes kept the controller
boxes 75% busy during backups after midnight for an hour, 30-40% busy
during daytime hours, and about 20% busy evenings during batch
processing. The 14 RS/6000s got the control units up to 8-9% busy
during backups just after midnight, otherwise it was 0, 1, or on
occasion 2% busy during the day.

Post by Dave Wade
Desktop PC with local IDE disk might not keep up. Decent server with fibre
SAN with dedicated disk and no redundancy will blow 3090 i/o out of the
water. Lets face it the 3390Mod27 disk is usually just the same SAN disk
that is used on PC servers. If you don't believe me down load some of the
Exchange benchmarks from the Microsoft web site, or Notes benchmarks from
the IBM web site and look at the i/o they do. I know it's a "benchmark" but
several manufacturers have reported simulating serving over 30,000 mail
users on a PC whilst maintaining sub-second response times. Work out the I/O
required and compare to 3090.
Mind you, you do need to do things properly. Just because its possible, as
shown by the many mainframe sites that put their DASD on SAN, it does not
mean its done properly for PC servers. All too often the I/O system is
compromised by tuning for capacity over performance.
Mind you, getting a "balanced system" is important. I personally think one
reason the DEC Alpha chip achieved so little penetration of the PC server
market place was that at the time it was "too fast" for the I/O in use.
Moving to the Alpha from Pentium achieved very little because the disk
subsystems of the time couldn't keep up.
Dave

--
Mike A Schwab, Springfield IL USA http://geocities.
<http://geocities.com/maschwab/> com/maschwab/ for
software links

[Non-text portions of this message have been removed]





[Non-text portions of this message have been removed]

[Josef Sipek]

Post by Paul Raulerson
Oh, I wouldn't say that. AIX or Linux can and will put a load on a San,
but to be perfectly honest, despite the great specifications on the PCI and
PCI-X busses and their kin, I have not yet seen a non-mainframe size pSeries
or iSeries box able to engage the same kind of load a mainframe can. Note,
the very top of end of both p and I series machines are as close to being a
mainframe as makes no difference, at least in terms of I/O. They are very
different machines from the rack mounted servers most people think of for
pSeries, and the server size boxes most people think of for iSeries
machines.
Mainframes, in particular mainframes running z/OS are elegant. No fooling,
they do things very differently from the typical low end machines. Hercules,
running on a PC, proves that it is not all in the hardware though, as even
the legal OSs, like MVS, can put "a hurting" on the PC's CPU and I/O
capacity. I would love to see it running z/OS, or z/VM, both of which will
run, but are forbidden of course.
z/Linux running under Hercules running under Linux does a respectable job
too.
It isn't in the design parameters for Hercules, but I would love to see
Hercules load as the hypervisor on a PC, without the overhead of a full
Linux load. I am convinced that would produce some amazing results. But
without a modern day legal O/S, it is a futile dream to pursue.

Linux doesn't need to add much overhead. Few ideas one might want to try:

1) boot with init=/usr/bin/hercules or init=/bin/bash and exec hercules
(this would make the process schedule's job really easy - only
hercules process/threads & few kernel daemons)
2) use raw disk device files for dasd/tape storage
(this would eliminate the filesystem layer for hercules emulated
I/O; there would still be some I/O due to module loading, etc. but
once all that is in memory, it should be all good)
3) although the UI for hercules is neat, displaying things on the screen
slows down the process doing it - I'm not sure how to disable that
in hercules but there must be a way

Josef "Jeff" Sipek.

[Tim Pinkawa]

Post by Josef Sipek
3) although the UI for hercules is neat, displaying things on the screen
slows down the process doing it - I'm not sure how to disable that
in hercules but there must be a way

Run Hercules with the -d (as in daemon) option to disable the interface.

Tim

[Josef Sipek]

Post by Tim Pinkawa

Post by Josef Sipek
3) although the UI for hercules is neat, displaying things on the screen
slows down the process doing it - I'm not sure how to disable that
in hercules but there must be a way

Run Hercules with the -d (as in daemon) option to disable the interface.

Neat!

If someone suggests a good "benchmark" load I could try (along with exactly
how do I run it - I'm pretty new to VM and MVS), I might try it on one of
the boxes I have access to (2CPU 2.8GHz Xeon, 2GB RAM, U320 SCSI disk).
Well, I'd try it both ways (1) with full linux running, using the
filesystem, etc., and (2) just the kernel with raw disks for IO.

Josef "Jeff" Sipek.

[Dave Morton]

PC/3090 Benchmarks:
Check old RMF reports and other sources from 15 years ago to obtain
the stats for the 3090's workload - batch and online usage, average
number of users swapped in, average region size for each group,
average number of regions allocated, number of jobs run, tasks run,
pages printed, etc, etc.

TSO benchmark:
1. Attach 700 3270-terminals (or PCs with 3270 emulation) to your PC.
2. Get 700 experienced TSO users to utilize the terminals.
3. Measure the response times (from ENTER KEY to FULL DISPLAY on the
screen - you might need TSOMON or some other monitor to get these
more accurate response time stats for TSO, rather than RMF - the RT
period must not end before the screen is completely displayed for
the user to view it).

CICS benchmark:
Similar - probably hundreds or thousands of users.

Batch workload benchmark:
Similar - probably dozens of batch jobs running at any given time.

Printing reports, compiles, etc:
Similar - probably multiple printers - impact printers and laser
printers (3211s, 3800s, etc). Get the number and types of printers,
pages printed per hour, etc.

Run all 4 benchmarks ***SIMULTANEOUSLY*** to simulate the real world
in which the 3090 operated.

Measure online response times, and elapsed times for batch jobs and
printed reports.

Publish your results.

There's no need to hobble your PC by preventing I/O caching in
memory or on your HDD. Let your PC use every advantage it has!!

There's a need to run much LONGER benchmarks (many hours, not a few
seconds), much LARGER benchmarks (hundreds or thousands of jobs,
tasks, and online users), and SIMULTANEOUS activity (all benchmarks
must run simultaneously - the 3090 was very expensive, and
management usually didn't permit only 1 thing to be running on the
company's very expensive computer during working hours, I can assure
you!!).

Dave M

Post by Josef Sipek

Post by Tim Pinkawa

Post by Josef Sipek
3) although the UI for hercules is neat, displaying things on the screen
slows down the process doing it - I'm not sure how to disable that
in hercules but there must be a way

Run Hercules with the -d (as in daemon) option to disable the interface.

Neat!
If someone suggests a good "benchmark" load I could try (along

with exactly

Post by Josef Sipek
how do I run it - I'm pretty new to VM and MVS), I might try it on one of
the boxes I have access to (2CPU 2.8GHz Xeon, 2GB RAM, U320 SCSI disk).
Well, I'd try it both ways (1) with full linux running, using the
filesystem, etc., and (2) just the kernel with raw disks for IO.
Josef "Jeff" Sipek.
--
Generous programmers from around the world all join forces to help
you shoot yourself in the foot for free.

[Josef Sipek]

Post by Dave Morton
Check old RMF reports and other sources from 15 years ago to obtain
the stats for the 3090's workload - batch and online usage, average
number of users swapped in, average region size for each group,
average number of regions allocated, number of jobs run, tasks run,
pages printed, etc, etc.
1. Attach 700 3270-terminals (or PCs with 3270 emulation) to your PC.
2. Get 700 experienced TSO users to utilize the terminals.

Ok, that'd be a show stopper :) Something more...automated would be nice.

Post by Dave Morton
3. Measure the response times (from ENTER KEY to FULL DISPLAY on the
screen - you might need TSOMON or some other monitor to get these
more accurate response time stats for TSO, rather than RMF - the RT
period must not end before the screen is completely displayed for
the user to view it).
Similar - probably hundreds or thousands of users.

Ditto.

Post by Dave Morton
Similar - probably dozens of batch jobs running at any given time.
Similar - probably multiple printers - impact printers and laser
printers (3211s, 3800s, etc). Get the number and types of printers,
pages printed per hour, etc.
Run all 4 benchmarks ***SIMULTANEOUSLY*** to simulate the real world
in which the 3090 operated.

Yeouch. I don't really want to try a real 3090 workload when I asked
(because I know the PC wouldn't handle it).

Post by Dave Morton
Measure online response times, and elapsed times for batch jobs and
printed reports.
Publish your results.

It would really be interesting to see someone do all this. I'm pretty sure
the PC wouldn't be able to keep up. If there is a way to "replace" the
thousands of TSO/etc. real people users with something scripted, I'd be up
for setting things up and trying to see exactly when the PC gives up. (But
I'd need guidance as I mentioned before I'm very new to VM and MVS.)

Post by Dave Morton
There's no need to hobble your PC by preventing I/O caching in
memory or on your HDD. Let your PC use every advantage it has!!

Makes sense. Selecting the "perfect" filesystem for this workload may be a
challenge on its own. :) Does one go for throughput? Or is not loading up
the CPU more important?

Post by Dave Morton
There's a need to run much LONGER benchmarks (many hours, not a few
seconds),

Agreed. I tried my hardest not to rip into the original post - the fact that
the whole simulation ran in less than a second effectively invalidates any
results. There's just far too much going on the average desktop computer.

Post by Dave Morton
much LARGER benchmarks (hundreds or thousands of jobs,
tasks, and online users),

Agreed.

Post by Dave Morton
and SIMULTANEOUS activity (all benchmarks
must run simultaneously - the 3090 was very expensive, and
management usually didn't permit only 1 thing to be running on the
company's very expensive computer during working hours, I can assure
you!!).

Running everything at the same time would surely have nasty caching effects
in the PC's CPU cache - and quite possibly even the I/O generated would
force disk buffer cache flushes.

It would be interesting thing to do, and if it can be all scripted, I'd like
to try - scripting it also adds more "stability" to the results as they can
be reproduced more easily.

Josef "Jeff" Sipek.

[Dave Morton]

Jeff, seriously, forget it.

How would you simulate the action of multiple mainframe printers,
printing on real paper??? Printing can't be simulated - it can only
be performed. What good is a simulation of printing? "Benchmarker
Guy 1: Where are those reports? Benchmarker Guy 2: They don't exist
and never will. This is all pretend... BG1: So we're just pretending
to benchmark a PC against a 3090? BG2: That's correct. This PC
doesn't stand a snowflake's chance in hell of matching the power of
that 3090 - in printing capacity or anything, else except for CPU
speed and burst I/O speed."

How many PC printers would it take to print all the reports done on
that 3090 during prime shift?
3211 = 18 pages/minute, 132 columns wide.
3800 = 40 pages/minute (approx, early model, 1970s).

They weren't terribly fast, back in the olden days, but they kept
grinding away for hours and days and weeks....

HP Deskjet = ?? pages/minute.

Remember, that's a PC-attached printer - not a high speed, small
laser printer attached to a mainframe. I dunno. Maybe a few of them
could keep up, but I'd need to see them in action to believe it. My
opinion: Extremely doubtful that they could keep up if attached to a
PC using today's PC software.

Of course, not everything is printed. CICS and SPF (and SDSF, etc)
are excellent viewing tools and can eliminate much printing - but
not all. Sometimes printed reports are easier to use than online
media, can be copied on a copy machine, can include penciled notes,
trends can be more easily spotted, they can be used when the
computer is down, etc.

You have a time machine. You go back in time and decide to replace
the company's 3090 with your PC. "User: Where are my reports? Jeff:
We don't print reports anymore. User: Who is your boss?"

"User: Where is my paycheck? Jeff: We don't print paychecks
anymore. The check images are just flung off into the ether like
vapor. We don't know where they go. We just simulate printing now.
User: How long have you worked here, not counting today?"

Then there's the issue of TAPES and tape drives. Tapes are important
because they're portable and can serve as backup media. Government
regs also require some kind of historical documentation. Tapes are
ideal for that purpose. Naturally, a full benchmark would need to
include the reading and writing of MANY tapes during a 24-hour day,
and a month, a quarter, and a year! That's point number 5, added to
the original 4 benchmark items. No tape creation = failure of the PC
to equal the power of the 3090.

Then there's VSAM, IMS, and DB2...

Then there's RACF....

Your PC doesn't stand a chance. To prove that it's the equal of a
3090 you would have to actually DO everything I mentioned. Running a
few scripts on your PC to simulate a 3090 in a commercial shop is so
far removed from reality, I don't know what to say... When you
actually DO everything (simultaneously, of course), please let us
know. Anything less than that is like claiming your PC is an
airplane because you have MS Flight Simulator installed... Taken any
real trips with it, lately?

Dave M

Post by Josef Sipek

Post by Dave Morton
Check old RMF reports and other sources from 15 years ago to

obtain

Post by Josef Sipek

Post by Dave Morton
the stats for the 3090's workload - batch and online usage,

average

Post by Josef Sipek

Post by Dave Morton
number of users swapped in, average region size for each group,
average number of regions allocated, number of jobs run, tasks run,
pages printed, etc, etc.
1. Attach 700 3270-terminals (or PCs with 3270 emulation) to

your PC.

Post by Josef Sipek

Post by Dave Morton
2. Get 700 experienced TSO users to utilize the terminals.

Ok, that'd be a show stopper :) Something more...automated would be nice.

Post by Dave Morton
3. Measure the response times (from ENTER KEY to FULL DISPLAY on the
screen - you might need TSOMON or some other monitor to get

these

Post by Josef Sipek

Post by Dave Morton
more accurate response time stats for TSO, rather than RMF - the RT
period must not end before the screen is completely displayed for
the user to view it).
Similar - probably hundreds or thousands of users.

Ditto.

Post by Dave Morton
Similar - probably dozens of batch jobs running at any given

time.

Post by Josef Sipek

Post by Dave Morton
Similar - probably multiple printers - impact printers and laser
printers (3211s, 3800s, etc). Get the number and types of

printers,

Post by Josef Sipek

Post by Dave Morton
pages printed per hour, etc.
Run all 4 benchmarks ***SIMULTANEOUSLY*** to simulate the real world
in which the 3090 operated.

Yeouch. I don't really want to try a real 3090 workload when I

asked

Post by Josef Sipek
(because I know the PC wouldn't handle it).

Post by Dave Morton
Measure online response times, and elapsed times for batch jobs and
printed reports.
Publish your results.

It would really be interesting to see someone do all this. I'm

pretty sure

Post by Josef Sipek
the PC wouldn't be able to keep up. If there is a way to "replace" the
thousands of TSO/etc. real people users with something scripted, I'd be up
for setting things up and trying to see exactly when the PC gives up. (But
I'd need guidance as I mentioned before I'm very new to VM and

MVS.)

Post by Josef Sipek

Post by Dave Morton
There's no need to hobble your PC by preventing I/O caching in
memory or on your HDD. Let your PC use every advantage it has!!

Makes sense. Selecting the "perfect" filesystem for this workload may be a
challenge on its own. :) Does one go for throughput? Or is not

loading up

Post by Josef Sipek
the CPU more important?

Post by Dave Morton
There's a need to run much LONGER benchmarks (many hours, not a few
seconds),

Agreed. I tried my hardest not to rip into the original post - the fact that
the whole simulation ran in less than a second effectively

invalidates any

Post by Josef Sipek
results. There's just far too much going on the average desktop computer.

Post by Dave Morton
much LARGER benchmarks (hundreds or thousands of jobs,
tasks, and online users),

Agreed.

Post by Dave Morton
and SIMULTANEOUS activity (all benchmarks
must run simultaneously - the 3090 was very expensive, and
management usually didn't permit only 1 thing to be running on the
company's very expensive computer during working hours, I can assure
you!!).

Running everything at the same time would surely have nasty

caching effects

Post by Josef Sipek
in the PC's CPU cache - and quite possibly even the I/O generated would
force disk buffer cache flushes.
It would be interesting thing to do, and if it can be all

scripted, I'd like

Post by Josef Sipek
to try - scripting it also adds more "stability" to the results as they can
be reproduced more easily.
Josef "Jeff" Sipek.
--
I want a pet mainframe.

[Paul Raulerson]

Hey Dave - have to disagree with you on the printing thing. Why waste
precious mainframe cycles on driving printers? It is without a doubt, the
perfect kind of thing for a small machine to do, one with a lot of cycles to
burn. J



We manage all the printing, enterprise wide, as well as the creation of
massive numbers of PDF reports, from a system that runs on a little Sun
Microsystems box- QDirect from RSA in New York.

Usually works like a champ! The mainframe can send it down to the box as
fast as the box can take it, and then the RSA box distributes it to all the
printers around the enterprise.



Just think of it like an external smart programmable peripheral card for the
mainframe. <grin>



-Paul



From: hercules-390-***@public.gmane.org [mailto:hercules-390-***@public.gmane.org] On
Behalf Of Dave Morton
Sent: Saturday, June 16, 2007 9:39 PM
To: hercules-390-***@public.gmane.org
Subject: [hercules-390] Now somewhat OT Re: IBM 3090 Model 600J compared
with modern desktop PC's.




Jeff, seriously, forget it.

How would you simulate the action of multiple mainframe printers,
printing on real paper??? Printing can't be simulated - it can only
be performed. What good is a simulation of printing? "Benchmarker
Guy 1: Where are those reports? Benchmarker Guy 2: They don't exist
and never will. This is all pretend... BG1: So we're just pretending
to benchmark a PC against a 3090? BG2: That's correct. This PC
doesn't stand a snowflake's chance in hell of matching the power of
that 3090 - in printing capacity or anything, else except for CPU
speed and burst I/O speed."

How many PC printers would it take to print all the reports done on
that 3090 during prime shift?
3211 = 18 pages/minute, 132 columns wide.
3800 = 40 pages/minute (approx, early model, 1970s).

They weren't terribly fast, back in the olden days, but they kept
grinding away for hours and days and weeks....

HP Deskjet = ?? pages/minute.

Remember, that's a PC-attached printer - not a high speed, small
laser printer attached to a mainframe. I dunno. Maybe a few of them
could keep up, but I'd need to see them in action to believe it. My
opinion: Extremely doubtful that they could keep up if attached to a
PC using today's PC software.

Of course, not everything is printed. CICS and SPF (and SDSF, etc)
are excellent viewing tools and can eliminate much printing - but
not all. Sometimes printed reports are easier to use than online
media, can be copied on a copy machine, can include penciled notes,
trends can be more easily spotted, they can be used when the
computer is down, etc.

You have a time machine. You go back in time and decide to replace
the company's 3090 with your PC. "User: Where are my reports? Jeff:
We don't print reports anymore. User: Who is your boss?"

"User: Where is my paycheck? Jeff: We don't print paychecks
anymore. The check images are just flung off into the ether like
vapor. We don't know where they go. We just simulate printing now.
User: How long have you worked here, not counting today?"

Then there's the issue of TAPES and tape drives. Tapes are important
because they're portable and can serve as backup media. Government
regs also require some kind of historical documentation. Tapes are
ideal for that purpose. Naturally, a full benchmark would need to
include the reading and writing of MANY tapes during a 24-hour day,
and a month, a quarter, and a year! That's point number 5, added to
the original 4 benchmark items. No tape creation = failure of the PC
to equal the power of the 3090.

Then there's VSAM, IMS, and DB2...

Then there's RACF....

Your PC doesn't stand a chance. To prove that it's the equal of a
3090 you would have to actually DO everything I mentioned. Running a
few scripts on your PC to simulate a 3090 in a commercial shop is so
far removed from reality, I don't know what to say... When you
actually DO everything (simultaneously, of course), please let us
know. Anything less than that is like claiming your PC is an
airplane because you have MS Flight Simulator installed... Taken any
real trips with it, lately?

Dave M

Post by Josef Sipek

Post by Dave Morton
Check old RMF reports and other sources from 15 years ago to

obtain

Post by Josef Sipek

Post by Dave Morton
the stats for the 3090's workload - batch and online usage,

average

Post by Josef Sipek

Post by Dave Morton
number of users swapped in, average region size for each group,
average number of regions allocated, number of jobs run, tasks run,
pages printed, etc, etc.
1. Attach 700 3270-terminals (or PCs with 3270 emulation) to

your PC.

Post by Josef Sipek

Post by Dave Morton
2. Get 700 experienced TSO users to utilize the terminals.

Ok, that'd be a show stopper :) Something more...automated would be nice.

Post by Dave Morton
3. Measure the response times (from ENTER KEY to FULL DISPLAY on the
screen - you might need TSOMON or some other monitor to get

these

Post by Josef Sipek

Post by Dave Morton
more accurate response time stats for TSO, rather than RMF - the RT
period must not end before the screen is completely displayed for
the user to view it).
Similar - probably hundreds or thousands of users.

Ditto.

Post by Dave Morton
Similar - probably dozens of batch jobs running at any given

time.

Post by Josef Sipek

Post by Dave Morton
Similar - probably multiple printers - impact printers and laser
printers (3211s, 3800s, etc). Get the number and types of

printers,

Post by Josef Sipek

Post by Dave Morton
pages printed per hour, etc.
Run all 4 benchmarks ***SIMULTANEOUSLY*** to simulate the real world
in which the 3090 operated.

Yeouch. I don't really want to try a real 3090 workload when I

asked

Post by Josef Sipek
(because I know the PC wouldn't handle it).

Post by Dave Morton
Measure online response times, and elapsed times for batch jobs and
printed reports.
Publish your results.

It would really be interesting to see someone do all this. I'm

pretty sure

Post by Josef Sipek
the PC wouldn't be able to keep up. If there is a way to "replace" the
thousands of TSO/etc. real people users with something scripted, I'd be up
for setting things up and trying to see exactly when the PC gives up. (But
I'd need guidance as I mentioned before I'm very new to VM and

MVS.)

Post by Josef Sipek

Post by Dave Morton
There's no need to hobble your PC by preventing I/O caching in
memory or on your HDD. Let your PC use every advantage it has!!

Makes sense. Selecting the "perfect" filesystem for this workload may be a
challenge on its own. :) Does one go for throughput? Or is not

loading up

Post by Josef Sipek
the CPU more important?

Post by Dave Morton
There's a need to run much LONGER benchmarks (many hours, not a few
seconds),

Agreed. I tried my hardest not to rip into the original post - the fact that
the whole simulation ran in less than a second effectively

invalidates any

Post by Josef Sipek
results. There's just far too much going on the average desktop computer.

Post by Dave Morton
much LARGER benchmarks (hundreds or thousands of jobs,
tasks, and online users),

Agreed.

Post by Dave Morton
and SIMULTANEOUS activity (all benchmarks
must run simultaneously - the 3090 was very expensive, and
management usually didn't permit only 1 thing to be running on the
company's very expensive computer during working hours, I can assure
you!!).

Running everything at the same time would surely have nasty

caching effects

Post by Josef Sipek
in the PC's CPU cache - and quite possibly even the I/O generated would
force disk buffer cache flushes.
It would be interesting thing to do, and if it can be all

scripted, I'd like

Post by Josef Sipek
to try - scripting it also adds more "stability" to the results as they can
be reproduced more easily.
Josef "Jeff" Sipek.
--
I want a pet mainframe.

[Non-text portions of this message have been removed]

[Dave Morton]

Paul - Okay, that's fine.

Is the Sun Microsystems computer a PC? I was speaking only of PCs
being slow at printing - at least in my experience.

In the really olden days, I heard that some shops used their old
1401 computers to print reports sent from the System/360s. That was
before HASP and JES. They may have used tapes - I'm not sure.

It's all a bit surprising that modern computers would offload the
printing function since JES2 is so efficient with its chaining of
CCWs, etc, but whatever. They must see a need for it. Perhaps the
REAL need was for PDF document creation, and they primarily wanted
to offload that function...

Another advantage would be the separation of the *operation* of
printing which can be quite extensive and complex in larger shops.

Dave M

Post by Paul Raulerson
Hey Dave - have to disagree with you on the printing thing. Why waste
precious mainframe cycles on driving printers? It is without a doubt, the
perfect kind of thing for a small machine to do, one with a lot of cycles to
burn. J
We manage all the printing, enterprise wide, as well as the

creation of

Post by Paul Raulerson
massive numbers of PDF reports, from a system that runs on a

little Sun

Post by Paul Raulerson
Microsystems box- QDirect from RSA in New York.
Usually works like a champ! The mainframe can send it down to the box as
fast as the box can take it, and then the RSA box distributes it to all the
printers around the enterprise.
Just think of it like an external smart programmable peripheral card for the
mainframe. <grin>
-Paul
Behalf Of Dave Morton
Sent: Saturday, June 16, 2007 9:39 PM
Subject: [hercules-390] Now somewhat OT Re: IBM 3090 Model 600J compared
with modern desktop PC's.
Jeff, seriously, forget it.
How would you simulate the action of multiple mainframe printers,
printing on real paper??? Printing can't be simulated - it can

only

Post by Paul Raulerson
be performed. What good is a simulation of printing? "Benchmarker
Guy 1: Where are those reports? Benchmarker Guy 2: They don't

exist

Post by Paul Raulerson
and never will. This is all pretend... BG1: So we're just

pretending

Post by Paul Raulerson
to benchmark a PC against a 3090? BG2: That's correct. This PC
doesn't stand a snowflake's chance in hell of matching the power of
that 3090 - in printing capacity or anything, else except for CPU
speed and burst I/O speed."
How many PC printers would it take to print all the reports done on
that 3090 during prime shift?
3211 = 18 pages/minute, 132 columns wide.
3800 = 40 pages/minute (approx, early model, 1970s).
They weren't terribly fast, back in the olden days, but they kept
grinding away for hours and days and weeks....
HP Deskjet = ?? pages/minute.
Remember, that's a PC-attached printer - not a high speed, small
laser printer attached to a mainframe. I dunno. Maybe a few of

them

Post by Paul Raulerson
could keep up, but I'd need to see them in action to believe it. My
opinion: Extremely doubtful that they could keep up if attached to a
PC using today's PC software.
Of course, not everything is printed. CICS and SPF (and SDSF, etc)
are excellent viewing tools and can eliminate much printing - but
not all. Sometimes printed reports are easier to use than online
media, can be copied on a copy machine, can include penciled

notes,

Post by Paul Raulerson
trends can be more easily spotted, they can be used when the
computer is down, etc.
You have a time machine. You go back in time and decide to replace
the company's 3090 with your PC. "User: Where are my reports?
We don't print reports anymore. User: Who is your boss?"
"User: Where is my paycheck? Jeff: We don't print paychecks
anymore. The check images are just flung off into the ether like
vapor. We don't know where they go. We just simulate printing now.
User: How long have you worked here, not counting today?"
Then there's the issue of TAPES and tape drives. Tapes are

important

Post by Paul Raulerson
because they're portable and can serve as backup media. Government
regs also require some kind of historical documentation. Tapes are
ideal for that purpose. Naturally, a full benchmark would need to
include the reading and writing of MANY tapes during a 24-hour

day,

Post by Paul Raulerson
and a month, a quarter, and a year! That's point number 5, added to
the original 4 benchmark items. No tape creation = failure of the PC
to equal the power of the 3090.
Then there's VSAM, IMS, and DB2...
Then there's RACF....
Your PC doesn't stand a chance. To prove that it's the equal of a
3090 you would have to actually DO everything I mentioned. Running a
few scripts on your PC to simulate a 3090 in a commercial shop is so
far removed from reality, I don't know what to say... When you
actually DO everything (simultaneously, of course), please let us
know. Anything less than that is like claiming your PC is an
airplane because you have MS Flight Simulator installed... Taken any
real trips with it, lately?
Dave M

40yahoogroups.com>

Post by Paul Raulerson

Post by Josef Sipek

Post by Dave Morton
Check old RMF reports and other sources from 15 years ago to

obtain

Post by Josef Sipek

Post by Dave Morton
the stats for the 3090's workload - batch and online usage,

average

Post by Josef Sipek

Post by Dave Morton
number of users swapped in, average region size for each

group,

Post by Paul Raulerson

Post by Josef Sipek

Post by Dave Morton
average number of regions allocated, number of jobs run, tasks

run,

Post by Josef Sipek

Post by Dave Morton
pages printed, etc, etc.
1. Attach 700 3270-terminals (or PCs with 3270 emulation) to

your PC.

Post by Josef Sipek

Post by Dave Morton
2. Get 700 experienced TSO users to utilize the terminals.

Ok, that'd be a show stopper :) Something more...automated would

be nice.

Post by Josef Sipek

Post by Dave Morton
3. Measure the response times (from ENTER KEY to FULL DISPLAY

on

Post by Paul Raulerson
the

Post by Josef Sipek

Post by Dave Morton
screen - you might need TSOMON or some other monitor to get

these

Post by Josef Sipek

Post by Dave Morton
more accurate response time stats for TSO, rather than RMF -

the

Post by Paul Raulerson
RT

Post by Josef Sipek

Post by Dave Morton
period must not end before the screen is completely displayed

for

Post by Josef Sipek

Post by Dave Morton
the user to view it).
Similar - probably hundreds or thousands of users.

Ditto.

Post by Dave Morton
Similar - probably dozens of batch jobs running at any given

time.

Post by Josef Sipek

Post by Dave Morton
Similar - probably multiple printers - impact printers and

laser

Post by Paul Raulerson

Post by Josef Sipek

Post by Dave Morton
printers (3211s, 3800s, etc). Get the number and types of

printers,

Post by Josef Sipek

Post by Dave Morton
pages printed per hour, etc.
Run all 4 benchmarks ***SIMULTANEOUSLY*** to simulate the real

world

Post by Josef Sipek

Post by Dave Morton
in which the 3090 operated.

Yeouch. I don't really want to try a real 3090 workload when I

asked

Post by Josef Sipek
(because I know the PC wouldn't handle it).

Post by Dave Morton
Measure online response times, and elapsed times for batch

jobs

Post by Paul Raulerson
and

Post by Josef Sipek

Post by Dave Morton
printed reports.
Publish your results.

It would really be interesting to see someone do all this. I'm

pretty sure

Post by Josef Sipek
the PC wouldn't be able to keep up. If there is a way

to "replace"

Post by Paul Raulerson
the

Post by Josef Sipek
thousands of TSO/etc. real people users with something scripted,

I'd be up

Post by Josef Sipek
for setting things up and trying to see exactly when the PC

gives

Post by Paul Raulerson
up. (But

Post by Josef Sipek
I'd need guidance as I mentioned before I'm very new to VM and

MVS.)

Post by Josef Sipek

Post by Dave Morton
There's no need to hobble your PC by preventing I/O caching in
memory or on your HDD. Let your PC use every advantage it

has!!

Post by Paul Raulerson

Post by Josef Sipek
Makes sense. Selecting the "perfect" filesystem for this

workload

Post by Paul Raulerson
may be a

Post by Josef Sipek
challenge on its own. :) Does one go for throughput? Or is not

loading up

Post by Josef Sipek
the CPU more important?

Post by Dave Morton
There's a need to run much LONGER benchmarks (many hours, not

a

Post by Paul Raulerson
few

Post by Josef Sipek

Post by Dave Morton
seconds),

Agreed. I tried my hardest not to rip into the original post -

the

Post by Paul Raulerson
fact that

Post by Josef Sipek
the whole simulation ran in less than a second effectively

invalidates any

Post by Josef Sipek
results. There's just far too much going on the average desktop

computer.

Post by Josef Sipek

Post by Dave Morton
much LARGER benchmarks (hundreds or thousands of jobs,
tasks, and online users),

Agreed.

Post by Dave Morton
and SIMULTANEOUS activity (all benchmarks
must run simultaneously - the 3090 was very expensive, and
management usually didn't permit only 1 thing to be running on

the

Post by Josef Sipek

Post by Dave Morton
company's very expensive computer during working hours, I can

assure

Post by Josef Sipek

Post by Dave Morton
you!!).

Running everything at the same time would surely have nasty

caching effects

Post by Josef Sipek
in the PC's CPU cache - and quite possibly even the I/O

generated

Post by Paul Raulerson
would

Post by Josef Sipek
force disk buffer cache flushes.
It would be interesting thing to do, and if it can be all

scripted, I'd like

Post by Josef Sipek
to try - scripting it also adds more "stability" to the results

as

Post by Paul Raulerson
they can

Post by Josef Sipek
be reproduced more easily.
Josef "Jeff" Sipek.
--
I want a pet mainframe.

[Non-text portions of this message have been removed]

[Paul Raulerson]

Surprisingly, yes, the Sun Microsystems computer is a PC - and Intel PC to
be precise. It is just running Solaris instead of Windows. In fact, it will
be on an IBM LS20 blade soon. J



This really is an area where mainframes are sometimes not the optimal
solution; modern printers are almost all page printers, which require
significant processing to get the most out of. In terms of I/O, they are
all "low speed" peripherals to the mainframe, and it makes a whole lot of
sense to present the mainframe with something that looks more like one high
speed peripheral than dozens, or hundreds of little low speed peripherals.
All of which require constant attention just to do things like print a page.
The REAL need was to do just what I described above - the PDF production was
sort of like "low hanging fruit." A freebie if you will.



Now, even here, the data to print is being generated on the mainframe, for
the most part. The RSA box does do Infoprint-like things, such as merging a
simple data stream from the mainframe with a complex overlay or set of
overlays.



But as you so aptly point out, the operation of the printers is now off the
mainframe, which makes the PC crowd happy - job security I suppose. ;)





From: hercules-390-***@public.gmane.org [mailto:hercules-390-***@public.gmane.org] On
Behalf Of Dave Morton
Sent: Saturday, June 16, 2007 10:19 PM
To: hercules-390-***@public.gmane.org
Subject: [hercules-390] Now somewhat OT Re: IBM 3090 Model 600J compared
with modern desktop PC's.



Paul - Okay, that's fine.

Is the Sun Microsystems computer a PC? I was speaking only of PCs
being slow at printing - at least in my experience.

In the really olden days, I heard that some shops used their old
1401 computers to print reports sent from the System/360s. That was
before HASP and JES. They may have used tapes - I'm not sure.

It's all a bit surprising that modern computers would offload the
printing function since JES2 is so efficient with its chaining of
CCWs, etc, but whatever. They must see a need for it. Perhaps the
REAL need was for PDF document creation, and they primarily wanted
to offload that function...

Another advantage would be the separation of the *operation* of
printing which can be quite extensive and complex in larger shops.

Dave M

--- In hercules-390-***@public.gmane.org <mailto:hercules-390%40yahoogroups.com>
, Paul Raulerson

Post by Paul Raulerson
Hey Dave - have to disagree with you on the printing thing. Why waste
precious mainframe cycles on driving printers? It is without a

doubt, the

Post by Paul Raulerson
perfect kind of thing for a small machine to do, one with a lot of cycles to
burn. J
We manage all the printing, enterprise wide, as well as the

creation of

Post by Paul Raulerson
massive numbers of PDF reports, from a system that runs on a

little Sun

Post by Paul Raulerson
Microsystems box- QDirect from RSA in New York.
Usually works like a champ! The mainframe can send it down to the

box as

Post by Paul Raulerson
fast as the box can take it, and then the RSA box distributes it to all the
printers around the enterprise.
Just think of it like an external smart programmable peripheral card for the
mainframe. <grin>
-Paul

[mailto:hercules-

Post by Paul Raulerson
Behalf Of Dave Morton
Sent: Saturday, June 16, 2007 9:39 PM
Subject: [hercules-390] Now somewhat OT Re: IBM 3090 Model 600J compared
with modern desktop PC's.
Jeff, seriously, forget it.
How would you simulate the action of multiple mainframe printers,
printing on real paper??? Printing can't be simulated - it can

only

Post by Paul Raulerson
be performed. What good is a simulation of printing? "Benchmarker
Guy 1: Where are those reports? Benchmarker Guy 2: They don't

exist

Post by Paul Raulerson
and never will. This is all pretend... BG1: So we're just

pretending

Post by Paul Raulerson
to benchmark a PC against a 3090? BG2: That's correct. This PC
doesn't stand a snowflake's chance in hell of matching the power of
that 3090 - in printing capacity or anything, else except for CPU
speed and burst I/O speed."
How many PC printers would it take to print all the reports done on
that 3090 during prime shift?
3211 = 18 pages/minute, 132 columns wide.
3800 = 40 pages/minute (approx, early model, 1970s).
They weren't terribly fast, back in the olden days, but they kept
grinding away for hours and days and weeks....
HP Deskjet = ?? pages/minute.
Remember, that's a PC-attached printer - not a high speed, small
laser printer attached to a mainframe. I dunno. Maybe a few of

them

Post by Paul Raulerson
could keep up, but I'd need to see them in action to believe it. My
opinion: Extremely doubtful that they could keep up if attached to a
PC using today's PC software.
Of course, not everything is printed. CICS and SPF (and SDSF, etc)
are excellent viewing tools and can eliminate much printing - but
not all. Sometimes printed reports are easier to use than online
media, can be copied on a copy machine, can include penciled

notes,

Post by Paul Raulerson
trends can be more easily spotted, they can be used when the
computer is down, etc.
You have a time machine. You go back in time and decide to replace
the company's 3090 with your PC. "User: Where are my reports?
We don't print reports anymore. User: Who is your boss?"
"User: Where is my paycheck? Jeff: We don't print paychecks
anymore. The check images are just flung off into the ether like
vapor. We don't know where they go. We just simulate printing now.
User: How long have you worked here, not counting today?"
Then there's the issue of TAPES and tape drives. Tapes are

important

Post by Paul Raulerson
because they're portable and can serve as backup media. Government
regs also require some kind of historical documentation. Tapes are
ideal for that purpose. Naturally, a full benchmark would need to
include the reading and writing of MANY tapes during a 24-hour

day,

Post by Paul Raulerson
and a month, a quarter, and a year! That's point number 5, added to
the original 4 benchmark items. No tape creation = failure of the PC
to equal the power of the 3090.
Then there's VSAM, IMS, and DB2...
Then there's RACF....
Your PC doesn't stand a chance. To prove that it's the equal of a
3090 you would have to actually DO everything I mentioned. Running a
few scripts on your PC to simulate a 3090 in a commercial shop is so
far removed from reality, I don't know what to say... When you
actually DO everything (simultaneously, of course), please let us
know. Anything less than that is like claiming your PC is an
airplane because you have MS Flight Simulator installed... Taken any
real trips with it, lately?
Dave M

<mailto:hercules-390%40yahoogroups.com> <mailto:hercules-390%
40yahoogroups.com>

Post by Paul Raulerson

Post by Josef Sipek

Post by Dave Morton
Check old RMF reports and other sources from 15 years ago to

obtain

Post by Josef Sipek

Post by Dave Morton
the stats for the 3090's workload - batch and online usage,

average

Post by Josef Sipek

Post by Dave Morton
number of users swapped in, average region size for each

group,

Post by Paul Raulerson

Post by Josef Sipek

Post by Dave Morton
average number of regions allocated, number of jobs run, tasks

run,

Post by Josef Sipek

Post by Dave Morton
pages printed, etc, etc.
1. Attach 700 3270-terminals (or PCs with 3270 emulation) to

your PC.

Post by Josef Sipek

Post by Dave Morton
2. Get 700 experienced TSO users to utilize the terminals.

Ok, that'd be a show stopper :) Something more...automated would

be nice.

Post by Josef Sipek

Post by Dave Morton
3. Measure the response times (from ENTER KEY to FULL DISPLAY

on

Post by Paul Raulerson
the

Post by Josef Sipek

Post by Dave Morton
screen - you might need TSOMON or some other monitor to get

these

Post by Josef Sipek

Post by Dave Morton
more accurate response time stats for TSO, rather than RMF -

the

Post by Paul Raulerson
RT

Post by Josef Sipek

Post by Dave Morton
period must not end before the screen is completely displayed

for

Post by Josef Sipek

Post by Dave Morton
the user to view it).
Similar - probably hundreds or thousands of users.

Ditto.

Post by Dave Morton
Similar - probably dozens of batch jobs running at any given

time.

Post by Josef Sipek

Post by Dave Morton
Similar - probably multiple printers - impact printers and

laser

Post by Paul Raulerson

Post by Josef Sipek

Post by Dave Morton
printers (3211s, 3800s, etc). Get the number and types of

printers,

Post by Josef Sipek

Post by Dave Morton
pages printed per hour, etc.
Run all 4 benchmarks ***SIMULTANEOUSLY*** to simulate the real

world

Post by Josef Sipek

Post by Dave Morton
in which the 3090 operated.

Yeouch. I don't really want to try a real 3090 workload when I

asked

Post by Josef Sipek
(because I know the PC wouldn't handle it).

Post by Dave Morton
Measure online response times, and elapsed times for batch

jobs

Post by Paul Raulerson
and

Post by Josef Sipek

Post by Dave Morton
printed reports.
Publish your results.

It would really be interesting to see someone do all this. I'm

pretty sure

Post by Josef Sipek
the PC wouldn't be able to keep up. If there is a way

to "replace"

Post by Paul Raulerson
the

Post by Josef Sipek
thousands of TSO/etc. real people users with something scripted,

I'd be up

Post by Josef Sipek
for setting things up and trying to see exactly when the PC

gives

Post by Paul Raulerson
up. (But

Post by Josef Sipek
I'd need guidance as I mentioned before I'm very new to VM and

MVS.)

Post by Josef Sipek

Post by Dave Morton
There's no need to hobble your PC by preventing I/O caching in
memory or on your HDD. Let your PC use every advantage it

has!!

Post by Paul Raulerson

Post by Josef Sipek
Makes sense. Selecting the "perfect" filesystem for this

workload

Post by Paul Raulerson
may be a

Post by Josef Sipek
challenge on its own. :) Does one go for throughput? Or is not

loading up

Post by Josef Sipek
the CPU more important?

Post by Dave Morton
There's a need to run much LONGER benchmarks (many hours, not

a

Post by Paul Raulerson
few

Post by Josef Sipek

Post by Dave Morton
seconds),

Agreed. I tried my hardest not to rip into the original post -

the

Post by Paul Raulerson
fact that

Post by Josef Sipek
the whole simulation ran in less than a second effectively

invalidates any

Post by Josef Sipek
results. There's just far too much going on the average desktop

computer.

Post by Josef Sipek

Post by Dave Morton
much LARGER benchmarks (hundreds or thousands of jobs,
tasks, and online users),

Agreed.

Post by Dave Morton
and SIMULTANEOUS activity (all benchmarks
must run simultaneously - the 3090 was very expensive, and
management usually didn't permit only 1 thing to be running on

the

Post by Josef Sipek

Post by Dave Morton
company's very expensive computer during working hours, I can

assure

Post by Josef Sipek

Post by Dave Morton
you!!).

Running everything at the same time would surely have nasty

caching effects

Post by Josef Sipek
in the PC's CPU cache - and quite possibly even the I/O

generated

Post by Paul Raulerson
would

Post by Josef Sipek
force disk buffer cache flushes.
It would be interesting thing to do, and if it can be all

scripted, I'd like

Post by Josef Sipek
to try - scripting it also adds more "stability" to the results

as

Post by Paul Raulerson
they can

Post by Josef Sipek
be reproduced more easily.
Josef "Jeff" Sipek.
--
I want a pet mainframe.

[Non-text portions of this message have been removed]

[Non-text portions of this message have been removed]

[Dave Morton]

Thanks for the info, Paul.

Since the mainframe is still the creator of the pages being sent to
the Sun computer at your shop (for the most part), that isn't saying
that the Sun computer is entirely responsible for printing. It's
doing the EASY part: sending the formatted pages to the printers.
The mainframe is still doing the heavy lifting AND sending the
formatted pages to the Sun. The Sun computer offloads part of the
mainframe's work - like a channel or control unit.

If the mainframe sent raw, unformatted pages to the Sun, then you
could say the Sun is really fast at formatting and printing when
using page printers, etc, but you can't say that since it's not
doing it (for the most part).

I don't know why they don't just put more intelligence (CPU power)
into the page printers and feed them raw data - mainframe and PC
printers. Perhaps it's not doable.

At any rate, I see your point regarding hundreds of printers for the
mainframe to keep track of, and presumably the issue of JES2's high
efficiency is mute where page printers are concerned, much
processing being involved in formatting the page.

Dave M

Post by Paul Raulerson
Surprisingly, yes, the Sun Microsystems computer is a PC - and Intel PC to
be precise. It is just running Solaris instead of Windows. In

fact, it will

Post by Paul Raulerson
be on an IBM LS20 blade soon. J
This really is an area where mainframes are sometimes not the

optimal

Post by Paul Raulerson
solution; modern printers are almost all page printers, which

require

Post by Paul Raulerson
significant processing to get the most out of. In terms of I/O, they are
all "low speed" peripherals to the mainframe, and it makes a whole lot of
sense to present the mainframe with something that looks more like one high
speed peripheral than dozens, or hundreds of little low speed

peripherals.

Post by Paul Raulerson
All of which require constant attention just to do things like

print a page.

Post by Paul Raulerson
The REAL need was to do just what I described above - the PDF

production was

Post by Paul Raulerson
sort of like "low hanging fruit." A freebie if you will.
Now, even here, the data to print is being generated on the

mainframe, for

Post by Paul Raulerson
the most part. The RSA box does do Infoprint-like things, such as merging a
simple data stream from the mainframe with a complex overlay or set of
overlays.
But as you so aptly point out, the operation of the printers is now off the
mainframe, which makes the PC crowd happy - job security I

suppose. ;)

Post by Paul Raulerson
Behalf Of Dave Morton
Sent: Saturday, June 16, 2007 10:19 PM
Subject: [hercules-390] Now somewhat OT Re: IBM 3090 Model 600J compared
with modern desktop PC's.
Paul - Okay, that's fine.
Is the Sun Microsystems computer a PC? I was speaking only of PCs
being slow at printing - at least in my experience.
In the really olden days, I heard that some shops used their old
1401 computers to print reports sent from the System/360s. That was
before HASP and JES. They may have used tapes - I'm not sure.
It's all a bit surprising that modern computers would offload the
printing function since JES2 is so efficient with its chaining of
CCWs, etc, but whatever. They must see a need for it. Perhaps the
REAL need was for PDF document creation, and they primarily wanted
to offload that function...
Another advantage would be the separation of the *operation* of
printing which can be quite extensive and complex in larger shops.
Dave M

40yahoogroups.com>

Post by Paul Raulerson
, Paul Raulerson

Post by Paul Raulerson
Hey Dave - have to disagree with you on the printing thing. Why

waste

Post by Paul Raulerson
precious mainframe cycles on driving printers? It is without a

doubt, the

Post by Paul Raulerson
perfect kind of thing for a small machine to do, one with a lot

of

Post by Paul Raulerson
cycles to

Post by Paul Raulerson
burn. J
We manage all the printing, enterprise wide, as well as the

creation of

Post by Paul Raulerson
massive numbers of PDF reports, from a system that runs on a

little Sun

Post by Paul Raulerson
Microsystems box- QDirect from RSA in New York.
Usually works like a champ! The mainframe can send it down to the

box as

Post by Paul Raulerson
fast as the box can take it, and then the RSA box distributes it

to all the

Post by Paul Raulerson
printers around the enterprise.
Just think of it like an external smart programmable peripheral

card for the

Post by Paul Raulerson
mainframe. <grin>
-Paul

[Doug Fuerst]

Actually, the Sun Systems do not use Intel very much, they started
with, and prefer, SPARC processors. Sun was not initially, and was
dragged kicking and screaming, to Intel.

Doug

Surprisingly, yes, the Sun Microsystems computer is a PC - and Intel PC to
be precise. It is just running Solaris instead of Windows. In fact, it will
be on an IBM LS20 blade soon. J
This really is an area where mainframes are sometimes not the optimal
solution; modern printers are almost all page printers, which require
significant processing to get the most out of. In terms of I/O, they are
all "low speed" peripherals to the mainframe, and it makes a whole lot of
sense to present the mainframe with something that looks more like one high
speed peripheral than dozens, or hundreds of little low speed peripherals.
All of which require constant attention just to do things like print a page.
The REAL need was to do just what I described above - the PDF production was
sort of like "low hanging fruit." A freebie if you will.
Now, even here, the data to print is being generated on the mainframe, for
the most part. The RSA box does do Infoprint-like things, such as merging a
simple data stream from the mainframe with a complex overlay or set of
overlays.
But as you so aptly point out, the operation of the printers is now off the
mainframe, which makes the PC crowd happy - job security I suppose. ;)
Behalf Of Dave Morton
Sent: Saturday, June 16, 2007 10:19 PM
Subject: [hercules-390] Now somewhat OT Re: IBM 3090 Model 600J compared
with modern desktop PC's.
Paul - Okay, that's fine.
Is the Sun Microsystems computer a PC? I was speaking only of PCs
being slow at printing - at least in my experience.
In the really olden days, I heard that some shops used their old
1401 computers to print reports sent from the System/360s. That was
before HASP and JES. They may have used tapes - I'm not sure.
It's all a bit surprising that modern computers would offload the
printing function since JES2 is so efficient with its chaining of
CCWs, etc, but whatever. They must see a need for it. Perhaps the
REAL need was for PDF document creation, and they primarily wanted
to offload that function...
Another advantage would be the separation of the *operation* of
printing which can be quite extensive and complex in larger shops.
Dave M
--- In
<mailto:hercules-390%40yahoogroups.com>
, Paul Raulerson

Post by Paul Raulerson
Hey Dave - have to disagree with you on the printing thing. Why

waste

Post by Paul Raulerson
precious mainframe cycles on driving printers? It is without a

doubt, the

Post by Paul Raulerson
perfect kind of thing for a small machine to do, one with a lot of

cycles to

Post by Paul Raulerson
burn. J
We manage all the printing, enterprise wide, as well as the

creation of

Post by Paul Raulerson
massive numbers of PDF reports, from a system that runs on a

little Sun

Post by Paul Raulerson
Microsystems box- QDirect from RSA in New York.
Usually works like a champ! The mainframe can send it down to the

box as

Post by Paul Raulerson
fast as the box can take it, and then the RSA box distributes it

to all the

Post by Paul Raulerson
printers around the enterprise.
Just think of it like an external smart programmable peripheral

card for the

Post by Paul Raulerson
mainframe. <grin>
-Paul

<mailto:hercules-390%40yahoogroups.com>
[mailto:hercules-
<mailto:390%40yahoogroups.com> ] On

Post by Paul Raulerson
Behalf Of Dave Morton
Sent: Saturday, June 16, 2007 9:39 PM

<mailto:hercules-390%40yahoogroups.com>

Post by Paul Raulerson
Subject: [hercules-390] Now somewhat OT Re: IBM 3090 Model 600J

compared

Post by Paul Raulerson
with modern desktop PC's.
Jeff, seriously, forget it.
How would you simulate the action of multiple mainframe printers,
printing on real paper??? Printing can't be simulated - it can

only

Post by Paul Raulerson
be performed. What good is a simulation of printing? "Benchmarker
Guy 1: Where are those reports? Benchmarker Guy 2: They don't

exist

Post by Paul Raulerson
and never will. This is all pretend... BG1: So we're just

pretending

Post by Paul Raulerson
to benchmark a PC against a 3090? BG2: That's correct. This PC
doesn't stand a snowflake's chance in hell of matching the power

of

Post by Paul Raulerson
that 3090 - in printing capacity or anything, else except for CPU
speed and burst I/O speed."
How many PC printers would it take to print all the reports done

on

Post by Paul Raulerson
that 3090 during prime shift?
3211 = 18 pages/minute, 132 columns wide.
3800 = 40 pages/minute (approx, early model, 1970s).
They weren't terribly fast, back in the olden days, but they kept
grinding away for hours and days and weeks....
HP Deskjet = ?? pages/minute.
Remember, that's a PC-attached printer - not a high speed, small
laser printer attached to a mainframe. I dunno. Maybe a few of

them

Post by Paul Raulerson
could keep up, but I'd need to see them in action to believe it.

My

Post by Paul Raulerson
opinion: Extremely doubtful that they could keep up if attached to

a

Post by Paul Raulerson
PC using today's PC software.
Of course, not everything is printed. CICS and SPF (and SDSF, etc)
are excellent viewing tools and can eliminate much printing - but
not all. Sometimes printed reports are easier to use than online
media, can be copied on a copy machine, can include penciled

notes,

Post by Paul Raulerson
trends can be more easily spotted, they can be used when the
computer is down, etc.
You have a time machine. You go back in time and decide to replace
the company's 3090 with your PC. "User: Where are my reports?
We don't print reports anymore. User: Who is your boss?"
"User: Where is my paycheck? Jeff: We don't print paychecks
anymore. The check images are just flung off into the ether like
vapor. We don't know where they go. We just simulate printing now.
User: How long have you worked here, not counting today?"
Then there's the issue of TAPES and tape drives. Tapes are

important

Post by Paul Raulerson
because they're portable and can serve as backup media. Government
regs also require some kind of historical documentation. Tapes are
ideal for that purpose. Naturally, a full benchmark would need to
include the reading and writing of MANY tapes during a 24-hour

day,

Post by Paul Raulerson
and a month, a quarter, and a year! That's point number 5, added

to

Post by Paul Raulerson
the original 4 benchmark items. No tape creation = failure of the

PC

Post by Paul Raulerson
to equal the power of the 3090.
Then there's VSAM, IMS, and DB2...
Then there's RACF....
Your PC doesn't stand a chance. To prove that it's the equal of a
3090 you would have to actually DO everything I mentioned. Running

a

Post by Paul Raulerson
few scripts on your PC to simulate a 3090 in a commercial shop is

so

Post by Paul Raulerson
far removed from reality, I don't know what to say... When you
actually DO everything (simultaneously, of course), please let us
know. Anything less than that is like claiming your PC is an
airplane because you have MS Flight Simulator installed... Taken

any

Post by Paul Raulerson
real trips with it, lately?
Dave M

<mailto:hercules-390%40yahoogroups.com> <mailto:hercules-390%
40yahoogroups.com>

Post by Paul Raulerson

Post by Josef Sipek

Post by Dave Morton
Check old RMF reports and other sources from 15 years ago to

obtain

Post by Josef Sipek

Post by Dave Morton
the stats for the 3090's workload - batch and online usage,

average

Post by Josef Sipek

Post by Dave Morton
number of users swapped in, average region size for each

group,

Post by Paul Raulerson

Post by Josef Sipek

Post by Dave Morton
average number of regions allocated, number of jobs run, tasks

run,

Post by Josef Sipek

Post by Dave Morton
pages printed, etc, etc.
1. Attach 700 3270-terminals (or PCs with 3270 emulation) to

your PC.

Post by Josef Sipek

Post by Dave Morton
2. Get 700 experienced TSO users to utilize the terminals.

Ok, that'd be a show stopper :) Something more...automated would

be nice.

Post by Josef Sipek

Post by Dave Morton
3. Measure the response times (from ENTER KEY to FULL DISPLAY

on

Post by Paul Raulerson
the

Post by Josef Sipek

Post by Dave Morton
screen - you might need TSOMON or some other monitor to get

these

Post by Josef Sipek

Post by Dave Morton
more accurate response time stats for TSO, rather than RMF -

the

Post by Paul Raulerson
RT

Post by Josef Sipek

Post by Dave Morton
period must not end before the screen is completely displayed

for

Post by Josef Sipek

Post by Dave Morton
the user to view it).
Similar - probably hundreds or thousands of users.

Ditto.

Post by Dave Morton
Similar - probably dozens of batch jobs running at any given

time.

Post by Josef Sipek

Post by Dave Morton
Similar - probably multiple printers - impact printers and

laser

Post by Paul Raulerson

Post by Josef Sipek

Post by Dave Morton
printers (3211s, 3800s, etc). Get the number and types of

printers,

Post by Josef Sipek

Post by Dave Morton
pages printed per hour, etc.
Run all 4 benchmarks ***SIMULTANEOUSLY*** to simulate the real

world

Post by Josef Sipek

Post by Dave Morton
in which the 3090 operated.

Yeouch. I don't really want to try a real 3090 workload when I

asked

Post by Josef Sipek
(because I know the PC wouldn't handle it).

Post by Dave Morton
Measure online response times, and elapsed times for batch

jobs

Post by Paul Raulerson
and

Post by Josef Sipek

Post by Dave Morton
printed reports.
Publish your results.

It would really be interesting to see someone do all this. I'm

pretty sure

Post by Josef Sipek
the PC wouldn't be able to keep up. If there is a way

to "replace"

Post by Paul Raulerson
the

Post by Josef Sipek
thousands of TSO/etc. real people users with something scripted,

I'd be up

Post by Josef Sipek
for setting things up and trying to see exactly when the PC

gives

Post by Paul Raulerson
up. (But

Post by Josef Sipek
I'd need guidance as I mentioned before I'm very new to VM and

MVS.)

Post by Josef Sipek

Post by Dave Morton
There's no need to hobble your PC by preventing I/O caching in
memory or on your HDD. Let your PC use every advantage it

has!!

Post by Paul Raulerson

Post by Josef Sipek
Makes sense. Selecting the "perfect" filesystem for this

workload

Post by Paul Raulerson
may be a

Post by Josef Sipek
challenge on its own. :) Does one go for throughput? Or is not

loading up

Post by Josef Sipek
the CPU more important?

Post by Dave Morton
There's a need to run much LONGER benchmarks (many hours, not

a

Post by Paul Raulerson
few

Post by Josef Sipek

Post by Dave Morton
seconds),

Agreed. I tried my hardest not to rip into the original post -

the

Post by Paul Raulerson
fact that

Post by Josef Sipek
the whole simulation ran in less than a second effectively

invalidates any

Post by Josef Sipek
results. There's just far too much going on the average desktop

computer.

Post by Josef Sipek

Post by Dave Morton
much LARGER benchmarks (hundreds or thousands of jobs,
tasks, and online users),

Agreed.

Post by Dave Morton
and SIMULTANEOUS activity (all benchmarks
must run simultaneously - the 3090 was very expensive, and
management usually didn't permit only 1 thing to be running on

the

Post by Josef Sipek

Post by Dave Morton
company's very expensive computer during working hours, I can

assure

Post by Josef Sipek

Post by Dave Morton
you!!).

Running everything at the same time would surely have nasty

caching effects

Post by Josef Sipek
in the PC's CPU cache - and quite possibly even the I/O

generated

Post by Paul Raulerson
would

Post by Josef Sipek
force disk buffer cache flushes.
It would be interesting thing to do, and if it can be all

scripted, I'd like

Post by Josef Sipek
to try - scripting it also adds more "stability" to the results

as

Post by Paul Raulerson
they can

Post by Josef Sipek
be reproduced more easily.
Josef "Jeff" Sipek.
--
I want a pet mainframe.

[Non-text portions of this message have been removed]

[Non-text portions of this message have been removed]

Doug Fuerst
Consultant
BK Associates
Brooklyn, NY
[phone] (Office)
[phone] (Fax)
[phone] (Cell)
doug-***@public.gmane.org


[Non-text portions of this message have been removed]

[Paul Raulerson]

They are moving to Intel very quickly I am afraid. A large portion of the
Sun workstations sold now are powered by Intel.

SPARC is very much nicer, and I like PowerPC a WHOLE lot better, but . the
world goes as it will, not as we would have it.



-Paul



From: hercules-390-***@public.gmane.org [mailto:hercules-390-***@public.gmane.org] On
Behalf Of Doug Fuerst
Sent: Sunday, June 17, 2007 8:04 AM
To: hercules-390-***@public.gmane.org
Subject: RE: [hercules-390] Now somewhat OT Re: IBM 3090 Model 600J compared
with modern desktop PC's.



Actually, the Sun Systems do not use Intel very much, they started
with, and prefer, SPARC processors. Sun was not initially, and was
dragged kicking and screaming, to Intel.

Doug

Surprisingly, yes, the Sun Microsystems computer is a PC - and Intel PC to
be precise. It is just running Solaris instead of Windows. In fact, it will
be on an IBM LS20 blade soon. J
This really is an area where mainframes are sometimes not the optimal
solution; modern printers are almost all page printers, which require
significant processing to get the most out of. In terms of I/O, they are
all "low speed" peripherals to the mainframe, and it makes a whole lot of
sense to present the mainframe with something that looks more like one high
speed peripheral than dozens, or hundreds of little low speed peripherals.
All of which require constant attention just to do things like print a page.
The REAL need was to do just what I described above - the PDF production was
sort of like "low hanging fruit." A freebie if you will.
Now, even here, the data to print is being generated on the mainframe, for
the most part. The RSA box does do Infoprint-like things, such as merging a
simple data stream from the mainframe with a complex overlay or set of
overlays.
But as you so aptly point out, the operation of the printers is now off the
mainframe, which makes the PC crowd happy - job security I suppose. ;)

<mailto:hercules-390%40yahoogroups.com>
<mailto:hercules-390%40yahoogroups.com> ] On

Behalf Of Dave Morton
Sent: Saturday, June 16, 2007 10:19 PM

<mailto:hercules-390%40yahoogroups.com>

Subject: [hercules-390] Now somewhat OT Re: IBM 3090 Model 600J compared
with modern desktop PC's.
Paul - Okay, that's fine.
Is the Sun Microsystems computer a PC? I was speaking only of PCs
being slow at printing - at least in my experience.
In the really olden days, I heard that some shops used their old
1401 computers to print reports sent from the System/360s. That was
before HASP and JES. They may have used tapes - I'm not sure.
It's all a bit surprising that modern computers would offload the
printing function since JES2 is so efficient with its chaining of
CCWs, etc, but whatever. They must see a need for it. Perhaps the
REAL need was for PDF document creation, and they primarily wanted
to offload that function...
Another advantage would be the separation of the *operation* of
printing which can be quite extensive and complex in larger shops.
Dave M
--- In

<mailto:hercules-390%40yahoogroups.com>

<mailto:hercules-390%40yahoogroups.com>
, Paul Raulerson

Post by Paul Raulerson
Hey Dave - have to disagree with you on the printing thing. Why

waste

Post by Paul Raulerson
precious mainframe cycles on driving printers? It is without a

doubt, the

Post by Paul Raulerson
perfect kind of thing for a small machine to do, one with a lot of

cycles to

Post by Paul Raulerson
burn. J
We manage all the printing, enterprise wide, as well as the

creation of

Post by Paul Raulerson
massive numbers of PDF reports, from a system that runs on a

little Sun

Post by Paul Raulerson
Microsystems box- QDirect from RSA in New York.
Usually works like a champ! The mainframe can send it down to the

box as

Post by Paul Raulerson
fast as the box can take it, and then the RSA box distributes it

to all the

Post by Paul Raulerson
printers around the enterprise.
Just think of it like an external smart programmable peripheral

card for the

Post by Paul Raulerson
mainframe. <grin>
-Paul

<mailto:hercules-390%40yahoogroups.com>

<mailto:hercules-390%40yahoogroups.com>
[mailto:hercules-

<mailto:390%40yahoogroups.com>

<mailto:390%40yahoogroups.com> ] On

Post by Paul Raulerson
Behalf Of Dave Morton
Sent: Saturday, June 16, 2007 9:39 PM

<mailto:hercules-390%40yahoogroups.com>

<mailto:hercules-390%40yahoogroups.com>

Post by Paul Raulerson
Subject: [hercules-390] Now somewhat OT Re: IBM 3090 Model 600J

compared

Post by Paul Raulerson
with modern desktop PC's.
Jeff, seriously, forget it.
How would you simulate the action of multiple mainframe printers,
printing on real paper??? Printing can't be simulated - it can

only

Post by Paul Raulerson
be performed. What good is a simulation of printing? "Benchmarker
Guy 1: Where are those reports? Benchmarker Guy 2: They don't

exist

Post by Paul Raulerson
and never will. This is all pretend... BG1: So we're just

pretending

Post by Paul Raulerson
to benchmark a PC against a 3090? BG2: That's correct. This PC
doesn't stand a snowflake's chance in hell of matching the power

of

Post by Paul Raulerson
that 3090 - in printing capacity or anything, else except for CPU
speed and burst I/O speed."
How many PC printers would it take to print all the reports done

on

Post by Paul Raulerson
that 3090 during prime shift?
3211 = 18 pages/minute, 132 columns wide.
3800 = 40 pages/minute (approx, early model, 1970s).
They weren't terribly fast, back in the olden days, but they kept
grinding away for hours and days and weeks....
HP Deskjet = ?? pages/minute.
Remember, that's a PC-attached printer - not a high speed, small
laser printer attached to a mainframe. I dunno. Maybe a few of

them

Post by Paul Raulerson
could keep up, but I'd need to see them in action to believe it.

My

Post by Paul Raulerson
opinion: Extremely doubtful that they could keep up if attached to

a

Post by Paul Raulerson
PC using today's PC software.
Of course, not everything is printed. CICS and SPF (and SDSF, etc)
are excellent viewing tools and can eliminate much printing - but
not all. Sometimes printed reports are easier to use than online
media, can be copied on a copy machine, can include penciled

notes,

Post by Paul Raulerson
trends can be more easily spotted, they can be used when the
computer is down, etc.
You have a time machine. You go back in time and decide to replace
the company's 3090 with your PC. "User: Where are my reports?
We don't print reports anymore. User: Who is your boss?"
"User: Where is my paycheck? Jeff: We don't print paychecks
anymore. The check images are just flung off into the ether like
vapor. We don't know where they go. We just simulate printing now.
User: How long have you worked here, not counting today?"
Then there's the issue of TAPES and tape drives. Tapes are

important

Post by Paul Raulerson
because they're portable and can serve as backup media. Government
regs also require some kind of historical documentation. Tapes are
ideal for that purpose. Naturally, a full benchmark would need to
include the reading and writing of MANY tapes during a 24-hour

day,

Post by Paul Raulerson
and a month, a quarter, and a year! That's point number 5, added

to

Post by Paul Raulerson
the original 4 benchmark items. No tape creation = failure of the

PC

Post by Paul Raulerson
to equal the power of the 3090.
Then there's VSAM, IMS, and DB2...
Then there's RACF....
Your PC doesn't stand a chance. To prove that it's the equal of a
3090 you would have to actually DO everything I mentioned. Running

a

Post by Paul Raulerson
few scripts on your PC to simulate a 3090 in a commercial shop is

so

Post by Paul Raulerson
far removed from reality, I don't know what to say... When you
actually DO everything (simultaneously, of course), please let us
know. Anything less than that is like claiming your PC is an
airplane because you have MS Flight Simulator installed... Taken

any

Post by Paul Raulerson
real trips with it, lately?
Dave M
--- In

<mailto:hercules-390%40yahoogroups.com>hercules-390-***@public.gmane.org
<mailto:hercules-390%40yahoogroups.com>

<mailto:hercules-390%40yahoogroups.com> <mailto:hercules-390%
40yahoogroups.com>

Post by Paul Raulerson

Post by Josef Sipek

Post by Dave Morton
Check old RMF reports and other sources from 15 years ago to

obtain

Post by Josef Sipek

Post by Dave Morton
the stats for the 3090's workload - batch and online usage,

average

Post by Josef Sipek

Post by Dave Morton
number of users swapped in, average region size for each

group,

Post by Paul Raulerson

Post by Josef Sipek

Post by Dave Morton
average number of regions allocated, number of jobs run, tasks

run,

Post by Josef Sipek

Post by Dave Morton
pages printed, etc, etc.
1. Attach 700 3270-terminals (or PCs with 3270 emulation) to

your PC.

Post by Josef Sipek

Post by Dave Morton
2. Get 700 experienced TSO users to utilize the terminals.

Ok, that'd be a show stopper :) Something more...automated would

be nice.

Post by Josef Sipek

Post by Dave Morton
3. Measure the response times (from ENTER KEY to FULL DISPLAY

on

Post by Paul Raulerson
the

Post by Josef Sipek

Post by Dave Morton
screen - you might need TSOMON or some other monitor to get

these

Post by Josef Sipek

Post by Dave Morton
more accurate response time stats for TSO, rather than RMF -

the

Post by Paul Raulerson
RT

Post by Josef Sipek

Post by Dave Morton
period must not end before the screen is completely displayed

for

Post by Josef Sipek

Post by Dave Morton
the user to view it).
Similar - probably hundreds or thousands of users.

Ditto.

Post by Dave Morton
Similar - probably dozens of batch jobs running at any given

time.

Post by Josef Sipek

Post by Dave Morton
Similar - probably multiple printers - impact printers and

laser

Post by Paul Raulerson

Post by Josef Sipek

Post by Dave Morton
printers (3211s, 3800s, etc). Get the number and types of

printers,

Post by Josef Sipek

Post by Dave Morton
pages printed per hour, etc.
Run all 4 benchmarks ***SIMULTANEOUSLY*** to simulate the real

world

Post by Josef Sipek

Post by Dave Morton
in which the 3090 operated.

Yeouch. I don't really want to try a real 3090 workload when I

asked

Post by Josef Sipek
(because I know the PC wouldn't handle it).

Post by Dave Morton
Measure online response times, and elapsed times for batch

jobs

Post by Paul Raulerson
and

Post by Josef Sipek

Post by Dave Morton
printed reports.
Publish your results.

It would really be interesting to see someone do all this. I'm

pretty sure

Post by Josef Sipek
the PC wouldn't be able to keep up. If there is a way

to "replace"

Post by Paul Raulerson
the

Post by Josef Sipek
thousands of TSO/etc. real people users with something scripted,

I'd be up

Post by Josef Sipek
for setting things up and trying to see exactly when the PC

gives

Post by Paul Raulerson
up. (But

Post by Josef Sipek
I'd need guidance as I mentioned before I'm very new to VM and

MVS.)

Post by Josef Sipek

Post by Dave Morton
There's no need to hobble your PC by preventing I/O caching in
memory or on your HDD. Let your PC use every advantage it

has!!

Post by Paul Raulerson

Post by Josef Sipek
Makes sense. Selecting the "perfect" filesystem for this

workload

Post by Paul Raulerson
may be a

Post by Josef Sipek
challenge on its own. :) Does one go for throughput? Or is not

loading up

Post by Josef Sipek
the CPU more important?

Post by Dave Morton
There's a need to run much LONGER benchmarks (many hours, not

a

Post by Paul Raulerson
few

Post by Josef Sipek

Post by Dave Morton
seconds),

Agreed. I tried my hardest not to rip into the original post -

the

Post by Paul Raulerson
fact that

Post by Josef Sipek
the whole simulation ran in less than a second effectively

invalidates any

Post by Josef Sipek
results. There's just far too much going on the average desktop

computer.

Post by Josef Sipek

Post by Dave Morton
much LARGER benchmarks (hundreds or thousands of jobs,
tasks, and online users),

Agreed.

Post by Dave Morton
and SIMULTANEOUS activity (all benchmarks
must run simultaneously - the 3090 was very expensive, and
management usually didn't permit only 1 thing to be running on

the

Post by Josef Sipek

Post by Dave Morton
company's very expensive computer during working hours, I can

assure

Post by Josef Sipek

Post by Dave Morton
you!!).

Running everything at the same time would surely have nasty

caching effects

Post by Josef Sipek
in the PC's CPU cache - and quite possibly even the I/O

generated

Post by Paul Raulerson
would

Post by Josef Sipek
force disk buffer cache flushes.
It would be interesting thing to do, and if it can be all

scripted, I'd like

Post by Josef Sipek
to try - scripting it also adds more "stability" to the results

as

Post by Paul Raulerson
they can

Post by Josef Sipek
be reproduced more easily.
Josef "Jeff" Sipek.
--
I want a pet mainframe.

[Non-text portions of this message have been removed]

[Non-text portions of this message have been removed]

Doug Fuerst
Consultant
BK Associates
Brooklyn, NY
[phone] (Office)
[phone] (Fax)
[phone] (Cell)
doug-***@public.gmane.org <mailto:doug%40bkassociates.net>

[Non-text portions of this message have been removed]





[Non-text portions of this message have been removed]

[jeffsavit]

These topics are always lots of fun, even if they're not particularly
meaningful: let's compare completely different machines designed for
different workloads as if "performance" was fungible, not to mention a
two-orders of magnitude performance hit for emulation!

A desktop PC's I/O not comparable with a 3090-600J because it can't do
bunches of CICS transactions or print? Well, the PC friend might say
the opposite is true because that 3090 couldn't run even a single GUI
user :-) Let's see: 1280x1024 displays with 16-bit color depth: how
many bytes do I transfer for a single repaint, and then how many can I
paint per second? A lot more transfer than a 4.5MB/s block mux could
transfer. And, let's not compare processes under Unix, Linux, or
Windows with CICS transactions, as it would be more fair to compare
them with complete address spaces, or to compare CICS to a web server.
That would be much fairer comparison of overhead matched to
transactions per second. But, we're doing this for fun, right?

As Gerhard illustrated, just change the parameters of the benchmark to
produce whatever result you're trying to get, and hope nobody figures
out what you're up to!

Some slightly more serious comments:

- desktop PCs have much more per-spindle disk bandwidth than the 3390s
of old, but typically have only one drive. I happen to have a Sun U40
workstation (dual CPU, dual core AMD) with 4 disk drives each >400GB.
I pretty much expect that I could outperform most 3090-era disk
configurations. Having 16GB RAM on the box means I don't have to go
back to disk for what I've read, either, at least for the kind of
workload I would have had 20 years ago - it would just live in RAM.
In any case: cached 10K RPM disks I buy today with 1/2 TB capacity
will beat the stuffing out of 4200 RPM drives we had years ago. If you
could evenly distribute your workload over many actuators that was
nice, but there were plenty of things (almost all the interesting
stuff) that serialized to a single device. VTOCs, PDS directories, etc.

- The game is *completely* different if you're talking about
server-class machines. I can't speak for POWER architecture, but I
know that there are AMD, Intel, and SPARC servers that do many
thousands of I/Os per second. When I joined Sun after many years as a
customer on mainframes, I thought I was the guy in the room that knew
all about high end I/O performance, and had to get a reality check.
There were 1998-era _medium size_ machines in production doing 9000
I/Os per second. We've since done tests on high end kit with over 1
million disk I/Os per second, and with over 18GB/second transfer.
Modern disk subsystems do everything that channels did for us in
offloading CPU work for I/O.

- It's also important to remember that performance - even CPU
performance - is "shaped" rather than being a single figure. Let's
look at that 3090-600 again. A single engine did about 20 MIPS, IIRC
(I actually ran on 3090-600S then, at 17MIPS). So, a single TCB -
essentially, a single CICS region - or a single CMS user could only
get one CPU's worth of power. I can easily outperform that with Hercules.

- To Paul's comment: more and more of our workstations are on AMD, and
soon Intel. For the desktop user, those chips are now where the market
lives, for both technical reasons (there's a lot of circuitry you need
to design around your chip, and single-process CPU performance at
lower cost is better on those chips than SPARC workstation chips), and
market reasons (the volume is in Intel and AMD, period). For the
server business, which is where Sun has made its hardware revenues for
many years - we stopped being primarily a workstation company a decade
ago - we're growing our Intel and AMD lines, but our SPARC product
lines are still the lion's share of our kit, and will probably
continue that way. Having our own architecture lets us do innovation
that is impossible in the commodity space. One of my lab machines is a
2-rack unit box with 32 virtual CPUs, drawing 100 watts. Neat huh? Sun
started, pre-SPARC. I wasnt' there then but IIRC it was on 68K class
machines and Intel 386. SPARC came later.

Sorry for long post - I've been lurking with comment for too long!

Other people argue over sports teams - computer people argue over
operating systems and hardware families. Good clean fun...

cheers, Jeff

Post by Paul Raulerson
They are moving to Intel very quickly I am afraid. A large portion of the
Sun workstations sold now are powered by Intel.
SPARC is very much nicer, and I like PowerPC a WHOLE lot better, but . the
world goes as it will, not as we would have it.

[Paul Raulerson]

LOL! Thanks Jeff - our little guys here is a Sparc, I don't know why I
thought he was Intel. Probably because the platform he is going to is x86.
I was just thinking of how to correct that error.



Actually, I used to run whole bunches (well 50 to 100) diskless
workstations of the 3/xxx generation, and was so happy when I got a Sun
4/260 to drive them with. I had TWO Fujitsu 1gb dirves and a 9-track tape
unit. Rebuilding the Kernel took only about 2 minutes, IIRC. I remember
feeling like I was swimming in MIPS!



It sure outpaced the IBM mainframe, the Sperry 1100, the Perkin Elmers, and
way outclassed my poor VAX 11/780. I learned right then and there that the
measure of a machine was NOT the MIPS rating. The only machines that really
kept up with the mainframe were our A-Series mainframe and that big old
4/260. Those A-Series machines are now hosted on Intel machines - the Unisys
Mainframe actually runs Windows. (*sigh*)



That same 4/260 machine would be dog slow today, even compared to the tiny
server we run. now if you guys would just put out some blades compatible
with an IBM BladeServer. In fact, Hercules on a good server, running OS/390
would probably outpace it in some ways. <grin>



-Paul



From: hercules-390-***@public.gmane.org [mailto:hercules-390-***@public.gmane.org] On
Behalf Of jeffsavit
Sent: Sunday, June 17, 2007 11:46 AM
To: hercules-390-***@public.gmane.org
Subject: [hercules-390] Now somewhat OT Re: IBM 3090 Model 600J compared
with modern desktop PC's.



These topics are always lots of fun, even if they're not particularly
meaningful: let's compare completely different machines designed for
different workloads as if "performance" was fungible, not to mention a
two-orders of magnitude performance hit for emulation!

A desktop PC's I/O not comparable with a 3090-600J because it can't do
bunches of CICS transactions or print? Well, the PC friend might say
the opposite is true because that 3090 couldn't run even a single GUI
user :-) Let's see: 1280x1024 displays with 16-bit color depth: how
many bytes do I transfer for a single repaint, and then how many can I
paint per second? A lot more transfer than a 4.5MB/s block mux could
transfer. And, let's not compare processes under Unix, Linux, or
Windows with CICS transactions, as it would be more fair to compare
them with complete address spaces, or to compare CICS to a web server.
That would be much fairer comparison of overhead matched to
transactions per second. But, we're doing this for fun, right?

As Gerhard illustrated, just change the parameters of the benchmark to
produce whatever result you're trying to get, and hope nobody figures
out what you're up to!

Some slightly more serious comments:

- desktop PCs have much more per-spindle disk bandwidth than the 3390s
of old, but typically have only one drive. I happen to have a Sun U40
workstation (dual CPU, dual core AMD) with 4 disk drives each >400GB.
I pretty much expect that I could outperform most 3090-era disk
configurations. Having 16GB RAM on the box means I don't have to go
back to disk for what I've read, either, at least for the kind of
workload I would have had 20 years ago - it would just live in RAM.
In any case: cached 10K RPM disks I buy today with 1/2 TB capacity
will beat the stuffing out of 4200 RPM drives we had years ago. If you
could evenly distribute your workload over many actuators that was
nice, but there were plenty of things (almost all the interesting
stuff) that serialized to a single device. VTOCs, PDS directories, etc.

- The game is *completely* different if you're talking about
server-class machines. I can't speak for POWER architecture, but I
know that there are AMD, Intel, and SPARC servers that do many
thousands of I/Os per second. When I joined Sun after many years as a
customer on mainframes, I thought I was the guy in the room that knew
all about high end I/O performance, and had to get a reality check.
There were 1998-era _medium size_ machines in production doing 9000
I/Os per second. We've since done tests on high end kit with over 1
million disk I/Os per second, and with over 18GB/second transfer.
Modern disk subsystems do everything that channels did for us in
offloading CPU work for I/O.

- It's also important to remember that performance - even CPU
performance - is "shaped" rather than being a single figure. Let's
look at that 3090-600 again. A single engine did about 20 MIPS, IIRC
(I actually ran on 3090-600S then, at 17MIPS). So, a single TCB -
essentially, a single CICS region - or a single CMS user could only
get one CPU's worth of power. I can easily outperform that with Hercules.

- To Paul's comment: more and more of our workstations are on AMD, and
soon Intel. For the desktop user, those chips are now where the market
lives, for both technical reasons (there's a lot of circuitry you need
to design around your chip, and single-process CPU performance at
lower cost is better on those chips than SPARC workstation chips), and
market reasons (the volume is in Intel and AMD, period). For the
server business, which is where Sun has made its hardware revenues for
many years - we stopped being primarily a workstation company a decade
ago - we're growing our Intel and AMD lines, but our SPARC product
lines are still the lion's share of our kit, and will probably
continue that way. Having our own architecture lets us do innovation
that is impossible in the commodity space. One of my lab machines is a
2-rack unit box with 32 virtual CPUs, drawing 100 watts. Neat huh? Sun
started, pre-SPARC. I wasnt' there then but IIRC it was on 68K class
machines and Intel 386. SPARC came later.

Sorry for long post - I've been lurking with comment for too long!

Other people argue over sports teams - computer people argue over
operating systems and hardware families. Good clean fun...

cheers, Jeff

--- In hercules-390-***@public.gmane.org <mailto:hercules-390%40yahoogroups.com>
, Paul Raulerson

Post by Paul Raulerson
They are moving to Intel very quickly I am afraid. A large portion of the
Sun workstations sold now are powered by Intel.
SPARC is very much nicer, and I like PowerPC a WHOLE lot better, but . the
world goes as it will, not as we would have it.

[Non-text portions of this message have been removed]

[jeffsavit]

Post by Paul Raulerson
now if you guys would just put out some blades compatible
with an IBM BladeServer. In fact, Hercules on a good server, running

OS/390 would probably outpace it in some ways. <grin>

http://www.sun.com/servers/index.jsp?cat=Sun%20Blade%20Servers&tab=3

<grin> back to you, too! :-)

PS: Now, puhleez - don't encourage me to do anything that can be
construed as marketing! Not that I think anyone is going to go out and
buy one of these guys because I posted here, but I want to make sure
we keep it light and in fun. Okay?

[peter_flass]

Post by jeffsavit
the opposite is true because that 3090 couldn't run even a single GUI
user :-) Let's see: 1280x1024 displays with 16-bit color depth: how
many bytes do I transfer for a single repaint, and then how many can I
paint per second? A lot more transfer than a 4.5MB/s block mux could
transfer.

To get a fair comparison you can't compre to the PC's display that has
a memory-mapped buffer. You'd have to look a X users - in this case
the number of X servers that the PC or mainframe X client could support.

Post by jeffsavit
And, let's not compare processes under Unix, Linux, or
Windows with CICS transactions, as it would be more fair to compare
them with complete address spaces, or to compare CICS to a web server.
That would be much fairer comparison of overhead matched to
transactions per second. But, we're doing this for fun, right?

I agree on the Web<=>CICS comparion, this occurred to me also. OTOH,
the overhead of a web server depends a lot on factors such as serving
static pages vs. running a CGI or servlet for each page. CICS is the
equivalent of all dynamic HTML. That's why I focused in on the TSO
users, which compare pretty exactly to PC or unix processes. I'm not
familiar with high-end peecee systems, but my guess is that 700
logged-in users would swamp most systems.

[jeffsavit]

Post by peter_flass
To get a fair comparison you can't compre to the PC's display that has
a memory-mapped buffer. You'd have to look a X users - in this case
the number of X servers that the PC or mainframe X client could support.

I agree. It's not an equal comparison to stack memory mapped I/O
against channel-based I/O. Philosophically, it shows you can have high
message-rate, low-latency I/O by mechanisms other than channel-based
architectures.

A test would be interesting to see. I know that at work we put dozens
of users on our "Sun Rays" (morally like an X terminal), on small
servers, but the protocol used is less chatty than X.

It would take some effort to drive simulated work loads, and of
course, we would want to make sure the constraining factor wasn't the
network, rather than the servers. Remember when we thought 9600 bits
per second was enough to serve a 3270 cluster's worth of users?

Which reminds me: does anyone have a copy of the "GDDM slot machine" demo?

Post by peter_flass
I agree on the Web<=>CICS comparion, this occurred to me also. OTOH,
the overhead of a web server depends a lot on factors such as serving
static pages vs. running a CGI or servlet for each page. CICS is the
equivalent of all dynamic HTML. That's why I focused in on the TSO
users, which compare pretty exactly to PC or unix processes. I'm not
familiar with high-end peecee systems, but my guess is that 700
logged-in users would swamp most systems.

It would also swamp a lot of mainframes of the day - let's recommend
something nicer and much less overhead, like CMS! (Jeff goes and puts
on his asbestos suit!) :)

[Dave Jones]

Hi, Jeff....

Post by jeffsavit
Which reminds me: does anyone have a copy of the "GDDM slot machine" demo?

I don't have that particular GDDM demo handy, but I do have a GDDM
version of 'Solitare', coded in PL/I.

A buddy of mine, Chris Langford (http://zfm.cestrian.com/) has recently
coded up both minesweeper and backgammon (the two player version) for
the 3270, using Rexx and running on CMS. They both work quite nicely, too.

[bass-jldX+]

Hi DJ,

I have a copy of the "GDDM slot machine" and Beetle.

Sam Bass

Post by Dave Jones
Hi, Jeff....

Post by jeffsavit
Which reminds me: does anyone have a copy of the "GDDM slot machine" demo?

I don't have that particular GDDM demo handy, but I do have a GDDM
version of 'Solitare', coded in PL/I.
A buddy of mine, Chris Langford (http://zfm.cestrian.com/) has recently
coded up both minesweeper and backgammon (the two player version) for
the 3270, using Rexx and running on CMS. They both work quite nicely, too.
--
DJ
V/Soft
http://groups.yahoo.com/group/hercules-390
http://www.hercules-390.org
Yahoo! Groups Links

[Dave Jones]

Hi, Sam.
Want to trade my Solitaire for your slot machine and beetle? I've got
botht the PL/I source coder and TEXT decks......

DJ

Post by bass-jldX+
Hi DJ,
I have a copy of the "GDDM slot machine" and Beetle.
Sam Bass

Post by Dave Jones
Hi, Jeff....

Post by jeffsavit
Which reminds me: does anyone have a copy of the "GDDM slot machine" demo?

I don't have that particular GDDM demo handy, but I do have a GDDM
version of 'Solitare', coded in PL/I.
A buddy of mine, Chris Langford (http://zfm.cestrian.com/) has recently
coded up both minesweeper and backgammon (the two player version) for
the 3270, using Rexx and running on CMS. They both work quite nicely, too.
--
DJ
V/Soft
http://groups.yahoo.com/group/hercules-390
http://www.hercules-390.org
Yahoo! Groups Links

[bass-jldX+]

Hi DJ,

Sure, my copies are at my web site.
www.kmbass.com/data/
Look for beetle.pli and fruit.pli
I think I might have commented a few things out (Includes) and copied it
inline so I could make some changes. Might have to tinker a bit with it.
I know they worked on MVS SP 1.3 or so, they were compiled with the PL/I
Optimizing compiler prior to 1984.

Sam

Post by Dave Jones
Hi, Sam.
Want to trade my Solitaire for your slot machine and beetle? I've got
botht the PL/I source coder and TEXT decks......
DJ

[jeffsavit]

Wow - thanks! I have no idea if I'll ever get it running under
Hercules (though I do have access to a z or two), but it would be so
cool.

Post by bass-jldX+
Hi DJ,
Sure, my copies are at my web site.
www.kmbass.com/data/
Look for beetle.pli and fruit.pli
I think I might have commented a few things out (Includes) and copied it
inline so I could make some changes. Might have to tinker a bit with it.
I know they worked on MVS SP 1.3 or so, they were compiled with the PL/I
Optimizing compiler prior to 1984.
Sam

Post by Dave Jones
Hi, Sam.
Want to trade my Solitaire for your slot machine and beetle? I've got
botht the PL/I source coder and TEXT decks......
DJ

[Chris Hedley]

Post by jeffsavit
It would take some effort to drive simulated work loads, and of
course, we would want to make sure the constraining factor wasn't the
network, rather than the servers. Remember when we thought 9600 bits
per second was enough to serve a 3270 cluster's worth of users?

I remember using a setup like that 15 or so years ago; we had a couple of
Unix systems doubling as "on-demand" mainframe terminal servers connected
through 9600 bit SDLC lines, which were usually fully utilised between 30
busy users and a couple of equally busy printers. They weren't fast, but
I guess I got kind of used to it after a couple of years. I was green
with envy when I visited the site where the mainframe lurked and saw what
the response time was like for those lucky enough to have channel-attached
terminals. I think our telecomms people were having some sort of
competition as to how many terminal sessions and phone calls they could
squeeze down a single 64kbit line.

But soon us Unix developers moved from our old VT220 compatibles onto
X-terminals and I decided I'd finally got the better part of the deal...

Chris.

[jeffsavit]

Yeah - it was shocking how much better local response time was
compared to snipping bits through a skinny straw.

Now, for a funny X story (well, I think it's funny, and I hope I
haven't told it here already). There was a manufacturing company that
had a bunch of X-terminals at the factory location, connected with
computers at HQ. They wanted to use the link for other stuff at
night, but it was completely congested - which made no sense at all,
because those users were day shift only. It should have been idle.

So, someone was dispatched to see what was going on: the computer room
with the terminals was near the factory floor, which had heavy
machinery generating vibration through the whole building. All the
mice were constantly shaking on their mice pads - you guessed it:
generating X events all night long as fast as they could be transmitted.

So, the instructions went out to the users: when you go home for the
night, turn the mouse over! (Turn the mouse into a turtle!)

cheers, Jeff

(snip) I was green
with envy when I visited the site where the mainframe lurked and saw what
the response time was like for those lucky enough to have

channel-attached

terminals. I think our telecomms people were having some sort of
competition as to how many terminal sessions and phone calls they could
squeeze down a single 64kbit line.
But soon us Unix developers moved from our old VT220 compatibles onto
X-terminals and I decided I'd finally got the better part of the deal...
Chris.

[Josef Sipek]

On Sun, Jun 17, 2007 at 02:39:17AM -0000, Dave Morton wrote:
...

Post by Dave Morton
"User: Where is my paycheck? Jeff: We don't print paychecks
anymore. The check images are just flung off into the ether like
vapor. We don't know where they go. We just simulate printing now.
User: How long have you worked here, not counting today?"

Um, isn't that called direct deposit? ;-)

Post by Dave Morton
Your PC doesn't stand a chance.

Agreed. The _full_ load of a 3090 would be enough of overwhelm any computer
short of a mainframe.

Post by Dave Morton
To prove that it's the equal of a 3090 you would have to actually DO
everything I mentioned.

That's not the point! The idea is NOT to replace a 3090 with a PC. The idea
is to see how much beating a PC running hercules could take.

Post by Dave Morton
Running a few scripts on your PC to simulate a 3090 in a commercial shop
is so far removed from reality, I don't know what to say... When you
actually DO everything (simultaneously, of course), please let us know.

Really the interesting part of the whole _benchmarking_ idea is to
progressively increase the load (the number of TSO users, batch jobs, etc.)
and see when the PC starts to fall behind. As all benchmarks, it wouldn't be
the real thing. (Benchmarks are by definition somewhat synthetic, no?)

Post by Dave Morton
Anything less than that is like claiming your PC is an airplane because
you have MS Flight Simulator installed...

No one is claiming anything like that. My PC (to my dismay) is not a
mainframe...

Josef "Jeff" Sipek.

[Dave Morton]

Post by wgs77
I remember the tour I took of the computer room, with the
enormous white boxes representing the CPU's of the 3090 Model 600J
and
also the 3380 and 3390 Disk Packs, it had to believe that my desktop
PC is superior. This is a numbing discovery for me, and its taking
some time for it to sink in...

This is why I challenged you.

Dave M

Post by wgs77
...

Post by Dave Morton
"User: Where is my paycheck? Jeff: We don't print paychecks
anymore. The check images are just flung off into the ether like
vapor. We don't know where they go. We just simulate printing now.
User: How long have you worked here, not counting today?"

Um, isn't that called direct deposit? ;-)

Post by Dave Morton
Your PC doesn't stand a chance.

Agreed. The _full_ load of a 3090 would be enough of overwhelm any computer
short of a mainframe.

Post by Dave Morton
To prove that it's the equal of a 3090 you would have to

actually DO

Post by wgs77

Post by Dave Morton
everything I mentioned.

That's not the point! The idea is NOT to replace a 3090 with a PC. The idea
is to see how much beating a PC running hercules could take.

Post by Dave Morton
Running a few scripts on your PC to simulate a 3090 in a

commercial shop

Post by wgs77

Post by Dave Morton
is so far removed from reality, I don't know what to say... When you
actually DO everything (simultaneously, of course), please let us know.

Really the interesting part of the whole _benchmarking_ idea is to
progressively increase the load (the number of TSO users, batch jobs, etc.)
and see when the PC starts to fall behind. As all benchmarks, it wouldn't be
the real thing. (Benchmarks are by definition somewhat synthetic, no?)

Post by Dave Morton
Anything less than that is like claiming your PC is an airplane because
you have MS Flight Simulator installed...

No one is claiming anything like that. My PC (to my dismay) is not a
mainframe...
Josef "Jeff" Sipek.
--
UNIX is user-friendly ... it's just selective about who it's

friends are

[Josef Sipek]

Post by Dave Morton

Post by wgs77
I remember the tour I took of the computer room, with the
enormous white boxes representing the CPU's of the 3090 Model 600J
and
also the 3380 and 3390 Disk Packs, it had to believe that my desktop
PC is superior. This is a numbing discovery for me, and its taking
some time for it to sink in...

This is why I challenged you.

I didn't say that. Bill did. :)

Josef "Jeff" Sipek.

[Paul Raulerson]

<grin> Well one of my PC's runs Hercules, and TOPS-20, and Linux all at the
same time. what else but a mainframe can do that? </>



Seriously, the P/390 was, without a doubt, a real mainframe. Hercules will
easily outpace and outperform one of those guys, except for printing I
suppose.

Has anyone taken on the challenge of getting printing working from Hercules?
I mean to a real printer, not to a file?

..

No one is claiming anything like that. My PC (to my dismay) is not a
mainframe...

Josef "Jeff" Sipek.






[Non-text portions of this message have been removed]

[Tim Pinkawa]

Post by Paul Raulerson
Has anyone taken on the challenge of getting printing working from Hercules?
I mean to a real printer, not to a file?

The print to pipe functionality in the Hercules emulated 1403 takes
all the work out of it. You just need to change your config file from

000E 1403 /path/to/output.txt

to

000E 1403 |lpr

assuming you're on a Linux host, that will pipe all the printed output
to lpr which, if properly configured, will go to your printer. This is
a trivial example and in reality you'd probably want to process it
with something like enscript for more readable print outs. I use a
similar trick to pipe to a shell script output processor I wrote which
generates mainframe-like landscape print outs in PDF format. It saves
a lot of paper while still producing more or less genuine output, and
if I really do need hard copy, I just print the PDF. In my case, by
changing the output destination of enscript, I could direct it to
always go to hard copy.

A more authentic option would be to use JRP (JES2 Remote Printers,
available on CBT Tape 249, file 158) and a pr3287 emulator.

Tim

[Paul Raulerson]

Does it handle splitting the jobs up? I will go give that a try right now, I
did not know (or remember) it was there.

I don't suppose you would be able to post that little script of yours, would
you? It sounds really like the ideal way to

Handle Herc printing to me.

-Paul



From: hercules-390-***@public.gmane.org [mailto:hercules-390-***@public.gmane.org] On
Behalf Of Tim Pinkawa
Sent: Sunday, June 17, 2007 12:14 AM
To: hercules-390-***@public.gmane.org
Subject: Re: [hercules-390] Now somewhat OT Re: IBM 3090 Model 600J compared
with modern desktop PC's.

Post by Paul Raulerson
Has anyone taken on the challenge of getting printing working from Hercules?
I mean to a real printer, not to a file?

The print to pipe functionality in the Hercules emulated 1403 takes
all the work out of it. You just need to change your config file from

000E 1403 /path/to/output.txt

to

000E 1403 |lpr

assuming you're on a Linux host, that will pipe all the printed output
to lpr which, if properly configured, will go to your printer. This is
a trivial example and in reality you'd probably want to process it
with something like enscript for more readable print outs. I use a
similar trick to pipe to a shell script output processor I wrote which
generates mainframe-like landscape print outs in PDF format. It saves
a lot of paper while still producing more or less genuine output, and
if I really do need hard copy, I just print the PDF. In my case, by
changing the output destination of enscript, I could direct it to
always go to hard copy.

A more authentic option would be to use JRP (JES2 Remote Printers,
available on CBT Tape 249, file 158) and a pr3287 emulator.

Tim





[Non-text portions of this message have been removed]

[Tim Pinkawa]

Post by Paul Raulerson
Does it handle splitting the jobs up? I will go give that a try right now, I
did not know (or remember) it was there.
I don't suppose you would be able to post that little script of yours, would
you? It sounds really like the ideal way to
Handle Herc printing to me.
-Paul

Yes, it handles splitting the jobs into separate files and then
optionally doing some sort of post-processing, like printing hard copy
or making PDFs. Your e-mail provided the motivation I needed to get it
all online. The main page is http://www.timpinkawa.net/hercules and
the specific page about printing is at
http://www.timpinkawa.net/hercules/prtspool.html

I'd appreciate any comments or feedback on the page and the program.

Thanks,
Tim

[Paul Raulerson]

This is really good stuff Tim - guess I know what I will be doing *this*
weekend! -Paul



From: hercules-390-***@public.gmane.org [mailto:hercules-390-***@public.gmane.org] On
Behalf Of Tim Pinkawa
Sent: Sunday, June 17, 2007 10:07 PM
To: hercules-390-***@public.gmane.org
Subject: Re: [hercules-390] Now somewhat OT Re: IBM 3090 Model 600J compared
with modern desktop PC's.

Post by Paul Raulerson
Does it handle splitting the jobs up? I will go give that a try right now, I
did not know (or remember) it was there.
I don't suppose you would be able to post that little script of yours, would
you? It sounds really like the ideal way to
Handle Herc printing to me.
-Paul

Yes, it handles splitting the jobs into separate files and then
optionally doing some sort of post-processing, like printing hard copy
or making PDFs. Your e-mail provided the motivation I needed to get it
all online. The main page is http://www.timpinkawa.net/hercules and
the specific page about printing is at
http://www.timpinkawa.net/hercules/prtspool.html

I'd appreciate any comments or feedback on the page and the program.

Thanks,
Tim





[Non-text portions of this message have been removed]

[Doug Fuerst]

And while the P390 certainly ran the 390 instruction set, it was NOT
a mainframe. If you have never seen or really worked on a real
mainframe, then you might not know the difference. My client's z/9's
and z/900's are a big bigger in scale than a P390. And the I/O is
exponentially larger.

Doug

Post by Paul Raulerson
<grin> Well one of my PC's runs Hercules, and TOPS-20, and Linux all at the
same time. what else but a mainframe can do that? </>
Seriously, the P/390 was, without a doubt, a real mainframe. Hercules will
easily outpace and outperform one of those guys, except for printing I
suppose.
Has anyone taken on the challenge of getting printing working from Hercules?
I mean to a real printer, not to a file?
..
No one is claiming anything like that. My PC (to my dismay) is not a
mainframe...
Josef "Jeff" Sipek.
[Non-text portions of this message have been removed]

Doug Fuerst
Consultant
BK Associates
Brooklyn, NY
[phone] (Office)
[phone] (Fax)
[phone] (Cell)
doug-***@public.gmane.org


[Non-text portions of this message have been removed]

[Binyamin Dissen]

On Sun, 17 Jun 2007 09:06:59 -0400 Doug Fuerst <doug-***@public.gmane.org> wrote:

:>And while the P390 certainly ran the 390 instruction set, it was NOT
:>a mainframe. If you have never seen or really worked on a real
:>mainframe, then you might not know the difference. My client's z/9's
:>and z/900's are a big bigger in scale than a P390. And the I/O is
:>exponentially larger.

The P390 did not have real channels. The I/O was done by the intel processor
running OS2.

[Jay Maynard]

Post by Binyamin Dissen
:>And while the P390 certainly ran the 390 instruction set, it was NOT
:>a mainframe. If you have never seen or really worked on a real
:>mainframe, then you might not know the difference. My client's z/9's
:>and z/900's are a big bigger in scale than a P390. And the I/O is
:>exponentially larger.
The P390 did not have real channels. The I/O was done by the intel processor
running OS2.

Not in that sense, no, but you could put a parallel channel adapter in one
and talk to channel attached devices with it. I've got a 3174-1L on one of
my P/390s.

[Gerhard Postpischil]

Post by Binyamin Dissen
The P390 did not have real channels. The I/O was done by the intel processor
running OS2.

Depends on what you mean by "real" channels. The P/390 came with
0-2 channel boards, that supported connection of gray and blue
Bus and Tag cables, and that was functionally very close to a
mainframe channel in behavior. It lacked the speed and capacity
for DASD support, but ran tapes, printers, 3174s, and similar gear.


Gerhard Postpischil
Bradford, VT

new e-mail address: gerhardp (at) charter (dot) net

[Gilson Lasco]

Post by Gerhard Postpischil
Depends on what you mean by "real" channels. The P/390 came with
0-2 channel boards, that supported connection of gray and blue Bus and Tag

cables, and >>that was functionally very close to a mainframe channel in
behavior. It lacked the >>

Post by Gerhard Postpischil
speed and capacity for DASD support, but ran tapes, printers, 3174s, and

similar gear.

Post by Gerhard Postpischil
Gerhard Postpischil

It also lacked the speed and capacity for the tape support.
I have a P390 with a channel adapter card and a couple of 3480 attached to
it,
and, as the OS/2 device drivers polls the 3480 C.U. ten times per second,
the best one can achieve is 16KB/s (when writing a tape file).
Dumping a full 3380-3 disc is a painful task ;-(

Gilson Lasco

[Paul Raulerson]

Actually, IBM sold it as a mainframe, it looked like a mainframe (even came
with a 20" monitor for the console) and software for it was priced like a
mainframe, until they came out with the now defunct ESL pricing. I know of
installations still using the p/390 in fact, though it is perhaps not cost
effective today given z9 pricing, zIIPs and zAAPs. In terms of 1994 pricing,
it was a miracle, and the RAID drives in the OS/2 or RS/6000 box it came in
were considered "blazingly fast" for the day. Laughable by today's standards
of course, but then, everything gets that way with time. Look at the fighter
aircraft from WWII for instance - hopeless outdated now, but state of the
art during their time!



I deal with a somewhat larger mainframe, though small by todays standards of
course. (A z800 2066.)



-Paul



From: hercules-390-***@public.gmane.org [mailto:hercules-390-***@public.gmane.org] On
Behalf Of Doug Fuerst
Sent: Sunday, June 17, 2007 8:07 AM
To: hercules-390-***@public.gmane.org
Subject: RE: [hercules-390] Now somewhat OT Re: IBM 3090 Model 600J compared
with modern desktop PC's.



And while the P390 certainly ran the 390 instruction set, it was NOT
a mainframe. If you have never seen or really worked on a real
mainframe, then you might not know the difference. My client's z/9's
and z/900's are a big bigger in scale than a P390. And the I/O is
exponentially larger.

Doug

Post by Paul Raulerson
<grin> Well one of my PC's runs Hercules, and TOPS-20, and Linux all at the
same time. what else but a mainframe can do that? </>
Seriously, the P/390 was, without a doubt, a real mainframe. Hercules will
easily outpace and outperform one of those guys, except for printing I
suppose.
Has anyone taken on the challenge of getting printing working from Hercules?
I mean to a real printer, not to a file?
..
No one is claiming anything like that. My PC (to my dismay) is not a
mainframe...
Josef "Jeff" Sipek.
[Non-text portions of this message have been removed]

Doug Fuerst
Consultant
BK Associates
Brooklyn, NY
[phone] (Office)
[phone] (Fax)
[phone] (Cell)
doug-***@public.gmane.org <mailto:doug%40bkassociates.net>

[Non-text portions of this message have been removed]





[Non-text portions of this message have been removed]

[Gerhard Postpischil]

Post by Doug Fuerst
And while the P390 certainly ran the 390 instruction set, it was NOT
a mainframe. If you have never seen or really worked on a real
mainframe, then you might not know the difference. My client's z/9's
and z/900's are a big bigger in scale than a P390. And the I/O is
exponentially larger.

It comes down to your definition of a mainframe. In the late
nineties I did consulting work for a small software house. They
had a P/390, with a channel attached string of 3480 drives, and
a 'real' line printer. Program development and testing for two
concurrent users was as responsive as any bigger machine I've
worked on.

I propose a variation on the Turing test - if you're sitting at
your terminal, can you tell what's on the other end, solely
based on your interaction with the machine?

Gerhard Postpischil
Bradford, VT

new e-mail address: gerhardp (at) charter (dot) net

[Martin Trübner]

Gerhard,

Post by Gerhard Postpischil
can you tell what's on the other end, solely

based on your interaction with the machine? <<

SO I am sitting on a TSO-terminal with my mainframe in the shade of a
willow-tree in my garden- excellent.

Can't tell if it is a 3090 or a 115 or a 67 or a 4361 or a z-109 ---it
just humms along

I think right now software believes its a 4361 (has power off, but no PR/SM)

[Adam Thornton]

Post by Paul Raulerson
<grin> Well one of my PC's runs Hercules, and TOPS-20, and Linux all at the
same time. what else but a mainframe can do that? </>

Where do I get TOPS-20 and an appropriate emulator? That sounds like
fun. SIMH and the Crispin TOPS-20 distro?

Adam

[Non-text portions of this message have been removed]

[Jay Maynard]

Post by Adam Thornton

Post by Paul Raulerson
<grin> Well one of my PC's runs Hercules, and TOPS-20, and Linux all at the
same time. what else but a mainframe can do that? </>

Where do I get TOPS-20 and an appropriate emulator? That sounds like
fun. SIMH and the Crispin TOPS-20 distro?

That seems to be the canonical TOPS-20 system these days...

[Mike Schwab]

http://pdp-10.trailing-edge.com/

Post by Adam Thornton
Where do I get TOPS-20 and an appropriate emulator? That sounds like
fun. SIMH and the Crispin TOPS-20 distro?
Adam

[Paul Raulerson]

http://panda.com/tops-20/



This will give you a really nice version of TOPS-20 v 7.1 or so, with all
the goodies loaded. Takes about 10 mins to setup, and there are pretty
decent instructions. It is a lot of fun - just like Herc!

-Paul



From: hercules-390-***@public.gmane.org [mailto:hercules-390-***@public.gmane.org] On
Behalf Of Adam Thornton
Sent: Sunday, June 17, 2007 12:24 PM
To: hercules-390-***@public.gmane.org
Subject: Re: [hercules-390] Now somewhat OT Re: IBM 3090 Model 600J compared
with modern desktop PC's.

Post by Paul Raulerson
<grin> Well one of my PC's runs Hercules, and TOPS-20, and Linux all at the
same time. what else but a mainframe can do that? </>

Where do I get TOPS-20 and an appropriate emulator? That sounds like
fun. SIMH and the Crispin TOPS-20 distro?

Adam

[Non-text portions of this message have been removed]





[Non-text portions of this message have been removed]

[Gerhard Postpischil]

Post by Josef Sipek
Really the interesting part of the whole _benchmarking_ idea is to
progressively increase the load (the number of TSO users, batch jobs, etc.)
and see when the PC starts to fall behind. As all benchmarks, it wouldn't be
the real thing. (Benchmarks are by definition somewhat synthetic, no?)

That brings back some fond memories. In the seventies our IBM
salesman tried to get us to replace a 360/65 with a 370/155, and
we actually installed one on a trial basis. My boss got a used
165, and asked me to run benchmarks. I asked him (facetiously)
which way he wanted the results to come out. There were enough
discrepancies in instruction timing for me to have made the 155
look better! In practice it could not support the 65's work
load. Back then we ran nearly a hundred Wylbur users with the
same system load as a dozen TSO sessions.

Gerhard Postpischil
Bradford, VT

new e-mail address: gerhardp (at) charter (dot) net

[Carey Tyler Schug]

Interesting comparison of the 65 and 155. I did benchmarks between IIRC
a 370/168 and a 3033. One interesting bit, of the various ways to clear
a moderate (e.g. 200 byte) area of storage:

XC field,field
MVC field, field-1 (assuming a zero in field -1)
MVC field,=nx'0'
LA 3,length/LA 2,field/XR 4,4/XR 5,5/MVCL 2,4
STM 15,13,savearea/LM 15,13,=14F'0'/STM 13,15,field, ( more as needed)/
LM 15,13,savearea (assuming 14 is base reg and area being cleared is
fullword aligned and a multiple of 4 bytes)

The above is IIRC, SLOWEST to FASTEST for a model 168 (which intuitively
makes sense), and FASTEST to SLOWEST (exactly reversed) for a 3033.
Clearly they added smarts in the microcode to recognize an XC with
identical addresses and just stuck zeros in memory without doing any
fetches or computation (XC field, field+1 was still incredibly slow).
Also, the overlapping MVC was optimized too. I think I checked and MVC
field, field-2 was MUCH SLOWER than the -1 version.

Post by Josef Sipek

Post by Josef Sipek
Really the interesting part of the whole _benchmarking_ idea is to
progressively increase the load (the number of TSO users, batch

jobs, etc.)

Post by Josef Sipek
and see when the PC starts to fall behind. As all benchmarks, it

wouldn't be

Post by Josef Sipek
the real thing. (Benchmarks are by definition somewhat synthetic, no?)

That brings back some fond memories. In the seventies our IBM
salesman tried to get us to replace a 360/65 with a 370/155, and
we actually installed one on a trial basis. My boss got a used
165, and asked me to run benchmarks. I asked him (facetiously)
which way he wanted the results to come out. There were enough
discrepancies in instruction timing for me to have made the 155
look better! In practice it could not support the 65's work
load. Back then we ran nearly a hundred Wylbur users with the
same system load as a dozen TSO sessions.
Gerhard Postpischil

[rhtatum]

Yes, governments do want historical records/archives. In fact, I know one company that built, among other things, high-pressure valves and archived all the test results to tape. The fact that they had some 30 year old tapes that were 7-track, BCD 556 BPI didn't faze tthem or the government folk that were supposed to monitor such things and assure (???) the public safety from exploding pipelines and booster stations ...:-)

But they had them in an off-site warehouse and paid good money for a non-climate-controlled facility there in Houston
--- I wondered, when I found out what they were doing, how many pieces of rotted acetate and oxide would fill the air if anyone ever retrieved a tape and opened the plastic case ...

Regards,
Ron Tatum
----- Original Message -----
From: Dave Morton
To: hercules-390-***@public.gmane.org
Sent: Saturday, June 16, 2007 9:39 PM
Subject: [hercules-390] Now somewhat OT Re: IBM 3090 Model 600J compared with modern desktop PC's.



Jeff, seriously, forget it.

How would you simulate the action of multiple mainframe printers,
printing on real paper??? Printing can't be simulated - it can only
be performed. What good is a simulation of printing? "Benchmarker
Guy 1: Where are those reports? Benchmarker Guy 2: They don't exist
and never will. This is all pretend... BG1: So we're just pretending
to benchmark a PC against a 3090? BG2: That's correct. This PC
doesn't stand a snowflake's chance in hell of matching the power of
that 3090 - in printing capacity or anything, else except for CPU
speed and burst I/O speed."

How many PC printers would it take to print all the reports done on
that 3090 during prime shift?
3211 = 18 pages/minute, 132 columns wide.
3800 = 40 pages/minute (approx, early model, 1970s).

They weren't terribly fast, back in the olden days, but they kept
grinding away for hours and days and weeks....

HP Deskjet = ?? pages/minute.

Remember, that's a PC-attached printer - not a high speed, small
laser printer attached to a mainframe. I dunno. Maybe a few of them
could keep up, but I'd need to see them in action to believe it. My
opinion: Extremely doubtful that they could keep up if attached to a
PC using today's PC software.

Of course, not everything is printed. CICS and SPF (and SDSF, etc)
are excellent viewing tools and can eliminate much printing - but
not all. Sometimes printed reports are easier to use than online
media, can be copied on a copy machine, can include penciled notes,
trends can be more easily spotted, they can be used when the
computer is down, etc.

You have a time machine. You go back in time and decide to replace
the company's 3090 with your PC. "User: Where are my reports? Jeff:
We don't print reports anymore. User: Who is your boss?"

"User: Where is my paycheck? Jeff: We don't print paychecks
anymore. The check images are just flung off into the ether like
vapor. We don't know where they go. We just simulate printing now.
User: How long have you worked here, not counting today?"

Then there's the issue of TAPES and tape drives. Tapes are important
because they're portable and can serve as backup media. Government
regs also require some kind of historical documentation. Tapes are
ideal for that purpose. Naturally, a full benchmark would need to
include the reading and writing of MANY tapes during a 24-hour day,
and a month, a quarter, and a year! That's point number 5, added to
the original 4 benchmark items. No tape creation = failure of the PC
to equal the power of the 3090.

Then there's VSAM, IMS, and DB2...

Then there's RACF....

Your PC doesn't stand a chance. To prove that it's the equal of a
3090 you would have to actually DO everything I mentioned. Running a
few scripts on your PC to simulate a 3090 in a commercial shop is so
far removed from reality, I don't know what to say... When you
actually DO everything (simultaneously, of course), please let us
know. Anything less than that is like claiming your PC is an
airplane because you have MS Flight Simulator installed... Taken any
real trips with it, lately?

Dave M

Post by Josef Sipek

Post by Dave Morton
Check old RMF reports and other sources from 15 years ago to

obtain

Post by Josef Sipek

Post by Dave Morton
the stats for the 3090's workload - batch and online usage,

average

Post by Josef Sipek

Post by Dave Morton
number of users swapped in, average region size for each group,
average number of regions allocated, number of jobs run, tasks run,
pages printed, etc, etc.
1. Attach 700 3270-terminals (or PCs with 3270 emulation) to

your PC.

Post by Josef Sipek

Post by Dave Morton
2. Get 700 experienced TSO users to utilize the terminals.

Ok, that'd be a show stopper :) Something more...automated would be nice.

Post by Dave Morton
3. Measure the response times (from ENTER KEY to FULL DISPLAY on the
screen - you might need TSOMON or some other monitor to get

these

Post by Josef Sipek

Post by Dave Morton
more accurate response time stats for TSO, rather than RMF - the RT
period must not end before the screen is completely displayed for
the user to view it).
Similar - probably hundreds or thousands of users.

Ditto.

Post by Dave Morton
Similar - probably dozens of batch jobs running at any given

time.

Post by Josef Sipek

Post by Dave Morton
Similar - probably multiple printers - impact printers and laser
printers (3211s, 3800s, etc). Get the number and types of

printers,

Post by Josef Sipek

Post by Dave Morton
pages printed per hour, etc.
Run all 4 benchmarks ***SIMULTANEOUSLY*** to simulate the real world
in which the 3090 operated.

Yeouch. I don't really want to try a real 3090 workload when I

asked

Post by Josef Sipek
(because I know the PC wouldn't handle it).

Post by Dave Morton
Measure online response times, and elapsed times for batch jobs and
printed reports.
Publish your results.

It would really be interesting to see someone do all this. I'm

pretty sure

Post by Josef Sipek
the PC wouldn't be able to keep up. If there is a way to "replace" the
thousands of TSO/etc. real people users with something scripted, I'd be up
for setting things up and trying to see exactly when the PC gives up. (But
I'd need guidance as I mentioned before I'm very new to VM and

MVS.)

Post by Josef Sipek

Post by Dave Morton
There's no need to hobble your PC by preventing I/O caching in
memory or on your HDD. Let your PC use every advantage it has!!

Makes sense. Selecting the "perfect" filesystem for this workload may be a
challenge on its own. :) Does one go for throughput? Or is not

loading up

Post by Josef Sipek
the CPU more important?

Post by Dave Morton
There's a need to run much LONGER benchmarks (many hours, not a few
seconds),

Agreed. I tried my hardest not to rip into the original post - the fact that
the whole simulation ran in less than a second effectively

invalidates any

Post by Josef Sipek
results. There's just far too much going on the average desktop computer.

Post by Dave Morton
much LARGER benchmarks (hundreds or thousands of jobs,
tasks, and online users),

Agreed.

Post by Dave Morton
and SIMULTANEOUS activity (all benchmarks
must run simultaneously - the 3090 was very expensive, and
management usually didn't permit only 1 thing to be running on the
company's very expensive computer during working hours, I can assure
you!!).

Running everything at the same time would surely have nasty

caching effects

Post by Josef Sipek
in the PC's CPU cache - and quite possibly even the I/O generated would
force disk buffer cache flushes.
It would be interesting thing to do, and if it can be all

scripted, I'd like

Post by Josef Sipek
to try - scripting it also adds more "stability" to the results as they can
be reproduced more easily.
Josef "Jeff" Sipek.
--
I want a pet mainframe.

[Non-text portions of this message have been removed]

[Dave Morton]

For 30-year data retention (if that was really necessary), I hope they
saved the hardcopy.

I was thinking in terms of 5-10 years since media technology changes
fairly rapidly, and tapes don't last forever.

However, there are viewable movies over 100 years old, although they
may be copies of the originals and may have been preserved in climate-
controlled conditions.

Dave M

Post by rhtatum
Yes, governments do want historical records/archives. In fact, I

know one company that built, among other things, high-pressure valves
and archived all the test results to tape. The fact that they had some
30 year old tapes that were 7-track, BCD 556 BPI didn't faze tthem or
the government folk that were supposed to monitor such things and
assure (???) the public safety from exploding pipelines and booster
stations ...:-)

Post by rhtatum
But they had them in an off-site warehouse and paid good money for a

non-climate-controlled facility there in Houston

Post by rhtatum
--- I wondered, when I found out what they were doing, how many

pieces of rotted acetate and oxide would fill the air if anyone ever
retrieved a tape and opened the plastic case ...

[Mike Schwab]

Didn't some members of the hercules groups sucessfully read some 30-40
year old 3420 tapes stored in attics?

Post by Dave Morton
For 30-year data retention (if that was really necessary), I hope they
saved the hardcopy.
I was thinking in terms of 5-10 years since media technology changes
fairly rapidly, and tapes don't last forever.
However, there are viewable movies over 100 years old, although they
may be copies of the originals and may have been preserved in climate-
controlled conditions.
Dave M

Post by rhtatum
Yes, governments do want historical records/archives. In fact, I

know one company that built, among other things, high-pressure valves
and archived all the test results to tape. The fact that they had some
30 year old tapes that were 7-track, BCD 556 BPI didn't faze tthem or
the government folk that were supposed to monitor such things and
assure (???) the public safety from exploding pipelines and booster
stations ...:-)

Post by rhtatum
But they had them in an off-site warehouse and paid good money for a

non-climate-controlled facility there in Houston

Post by rhtatum
--- I wondered, when I found out what they were doing, how many

pieces of rotted acetate and oxide would fill the air if anyone ever
retrieved a tape and opened the plastic case ...

[Rob van der Heij]

Post by Dave Morton
For 30-year data retention (if that was really necessary), I hope they
saved the hardcopy.

[more OT] Sigh. My parents had only very few pictures of when they
were young, simply because owning a camera was not common and my
grandparents had to visit a photographer shop to have a picture taken.
There's some pictures from my youth, and there's loads of pictures
from the next generation.

But I predict that the current generation of kids will look back
eventually and conclude they have no pictures from their youth either.
Almost everyone switched to digital images and consumers are not aware
of the complications to retain data for way more than a few years.
Quality of media like self-made CD's is probably less concern than
long-term compatibility (recently lost my last PC with 5.25" floppy,
so I disposed of most diskettes as well).

Some people do make hard copies of some of their digital images, but
the market for prints is mainly driven by price and I have no idea how
long those prints will last (in the past you also risked to lose your
film so you would be more selective).

I have some hope in Internet services like Flickr and Google's Picasa
to retain our past, as long as it remains interesting for
advertising...

Rob

[Dave Jones]

Josef Sipek wrote:
[snip....]

Post by Josef Sipek
It would really be interesting to see someone do all this. I'm pretty sure
the PC wouldn't be able to keep up. If there is a way to "replace" the
thousands of TSO/etc. real people users with something scripted, I'd be up
for setting things up and trying to see exactly when the PC gives up. (But
I'd need guidance as I mentioned before I'm very new to VM and MVS.)

For VM at least, IBM has made a script base test tool, called CHUG,
available on the VM download web page:

http://www.vm.ibm.com/download/packages/descript.cgi?CHUG

It might be a good place to start if folks are seriously interested in
stress testing VM running on Hercules.

[snip....]

Post by Josef Sipek
It would be interesting thing to do, and if it can be all scripted, I'd like
to try - scripting it also adds more "stability" to the results as they can
be reproduced more easily.
Josef "Jeff" Sipek.

A number of years ago (mid 1990s, maybe), a major Manhattan-based
brokerage house developed a VM-based Web application that was meant to
used by it's external customers. The Web server itself was being hosted
on VM, with the actual processing being done both of VM and on other
platforms (notably, z/OS and Unix system, as I recall). In order to
stress test the Web sever, the IT group wrote some test scripts that ran
on Unix boxes that made HTTP connections to the VM system and ran some
canned transactions.....the stress tests were a failure because the IT
group could not scrounge up enough spare Unix servers to create enough
of a load to cause the Web server to fail. The IT group had access to
*lots* of Unix servers, too. ;-)

[Dave Wade]

Dave,

Pity we don't have a modern VM to run it on but I was thinking of something
along those lines.



Jeff,

Well my experience is the reverse. I know the spool always was the weak
point of mainframes, but every Microsoft Exchange to SNADS/PROFS I have
installed has broken the mainframe, with the actual volume of data it can
send, and the Mainframes never had the storage to cope. These days I guess
it would be different if any one used Mainframe e-mail... and of course this
may also be a reflection about the garbage PC based e-mail systems send. but
never did the mainframe break my PC



Dave



-----Original Message-----
From: hercules-390-***@public.gmane.org [mailto:hercules-390-***@public.gmane.org] On
Behalf Of Dave Jones
Sent: 17 June 2007 16:44
To: hercules-390-***@public.gmane.org
Subject: Re: [hercules-390] Now somewhat OT Re: IBM 3090 Model 600J compared
with modern desktop PC's.





Josef Sipek wrote:
[snip....]

Post by Josef Sipek
It would really be interesting to see someone do all this. I'm pretty sure
the PC wouldn't be able to keep up. If there is a way to "replace" the
thousands of TSO/etc. real people users with something scripted, I'd be up
for setting things up and trying to see exactly when the PC gives up. (But
I'd need guidance as I mentioned before I'm very new to VM and MVS.)

For VM at least, IBM has made a script base test tool, called CHUG,
available on the VM download web page:

http://www.vm. <http://www.vm.ibm.com/download/packages/descript.cgi?CHUG>
ibm.com/download/packages/descript.cgi?CHUG

It might be a good place to start if folks are seriously interested in
stress testing VM running on Hercules.

[snip....]

Post by Josef Sipek
It would be interesting thing to do, and if it can be all scripted, I'd like
to try - scripting it also adds more "stability" to the results as they can
be reproduced more easily.
Josef "Jeff" Sipek.

A number of years ago (mid 1990s, maybe), a major Manhattan-based
brokerage house developed a VM-based Web application that was meant to
used by it's external customers. The Web server itself was being hosted
on VM, with the actual processing being done both of VM and on other
platforms (notably, z/OS and Unix system, as I recall). In order to
stress test the Web sever, the IT group wrote some test scripts that ran
on Unix boxes that made HTTP connections to the VM system and ran some
canned transactions.....the stress tests were a failure because the IT
group could not scrounge up enough spare Unix servers to create enough
of a load to cause the Web server to fail. The IT group had access to
*lots* of Unix servers, too. ;-)

[Paul Raulerson]

I don't think that runs on VM/370 Dave. more the pity. At least, I could
not get it to - it wanted stuff available only in newer versions.

-Paul



From: hercules-390-***@public.gmane.org [mailto:hercules-390-***@public.gmane.org] On
Behalf Of Dave Jones
Sent: Sunday, June 17, 2007 10:44 AM
To: hercules-390-***@public.gmane.org
Subject: Re: [hercules-390] Now somewhat OT Re: IBM 3090 Model 600J compared
with modern desktop PC's.





Josef Sipek wrote:
[snip....]

Post by Josef Sipek
It would really be interesting to see someone do all this. I'm pretty sure
the PC wouldn't be able to keep up. If there is a way to "replace" the
thousands of TSO/etc. real people users with something scripted, I'd be up
for setting things up and trying to see exactly when the PC gives up. (But
I'd need guidance as I mentioned before I'm very new to VM and MVS.)

For VM at least, IBM has made a script base test tool, called CHUG,
available on the VM download web page:

http://www.vm.ibm.com/download/packages/descript.cgi?CHUG

It might be a good place to start if folks are seriously interested in
stress testing VM running on Hercules.

[snip....]

Post by Josef Sipek
It would be interesting thing to do, and if it can be all scripted, I'd like
to try - scripting it also adds more "stability" to the results as they can
be reproduced more easily.
Josef "Jeff" Sipek.

A number of years ago (mid 1990s, maybe), a major Manhattan-based
brokerage house developed a VM-based Web application that was meant to
used by it's external customers. The Web server itself was being hosted
on VM, with the actual processing being done both of VM and on other
platforms (notably, z/OS and Unix system, as I recall). In order to
stress test the Web sever, the IT group wrote some test scripts that ran
on Unix boxes that made HTTP connections to the VM system and ran some
canned transactions.....the stress tests were a failure because the IT
group could not scrounge up enough spare Unix servers to create enough
of a load to cause the Web server to fail. The IT group had access to
*lots* of Unix servers, too. ;-)

[Dave Jones]

Well, poop...:-( I suppose one of the things missing in VM/370 is
support for LDEVs (logical devices), which CHUG needs. The doc mentions
that it works on VM/SP R5, which is around the 1988-1989 time frame.....

Happy Father's Day to all the fathers out there, as well.

Post by Paul Raulerson
I don't think that runs on VM/370 Dave. more the pity. At least, I could
not get it to - it wanted stuff available only in newer versions.
-Paul
Behalf Of Dave Jones
Sent: Sunday, June 17, 2007 10:44 AM
Subject: Re: [hercules-390] Now somewhat OT Re: IBM 3090 Model 600J compared
with modern desktop PC's.
[snip....]

Post by Josef Sipek
It would really be interesting to see someone do all this. I'm pretty sure
the PC wouldn't be able to keep up. If there is a way to "replace" the
thousands of TSO/etc. real people users with something scripted, I'd be up
for setting things up and trying to see exactly when the PC gives up. (But
I'd need guidance as I mentioned before I'm very new to VM and MVS.)

For VM at least, IBM has made a script base test tool, called CHUG,
http://www.vm.ibm.com/download/packages/descript.cgi?CHUG
It might be a good place to start if folks are seriously interested in
stress testing VM running on Hercules.
[snip....]

Post by Josef Sipek
It would be interesting thing to do, and if it can be all scripted, I'd

like

Post by Josef Sipek
to try - scripting it also adds more "stability" to the results as they

can

Post by Josef Sipek
be reproduced more easily.
Josef "Jeff" Sipek.

A number of years ago (mid 1990s, maybe), a major Manhattan-based
brokerage house developed a VM-based Web application that was meant to
used by it's external customers. The Web server itself was being hosted
on VM, with the actual processing being done both of VM and on other
platforms (notably, z/OS and Unix system, as I recall). In order to
stress test the Web sever, the IT group wrote some test scripts that ran
on Unix boxes that made HTTP connections to the VM system and ran some
canned transactions.....the stress tests were a failure because the IT
group could not scrounge up enough spare Unix servers to create enough
of a load to cause the Web server to fail. The IT group had access to
*lots* of Unix servers, too. ;-)

[Rob van der Heij]

Post by Sternbach, William
I came to some surprising conclusions after doing this research,
specifically at how slow and low capacity
this high end mainframe 3090 600J computer really was compared with
today's desktop PC's.

I'm afraid it's less of a surprise for me. There's probably 15 years
in between, and Moore says we double capacity every 18 months. So if
nothing else, there should be 3 order of magnitude difference between
a 3090 and a Pentium 4. I believe that's about what we see in the
table.

Even when you compare mainframe CPU's with Intel CPU's of the same
period, you'll find the mainframe CPU slower. The mainframe CPU has a
lot of extra stuff for error detection and recovery. That makes the
chip bigger and thus slower, given the same circuit technology. And
dual core on the mainframe means two cores doing the same instructions
and comparing their results.
Some environments can take advantage of doing it right rather than
fast (and sometimes wrong).

Rob

[Sternbach, William]

Hi,

I believe the reason we don't have 700 users connecting to our desktop
PC
is more a function of the high overhead of modern Windows Operating
Systems.
Desktop PC's today are thousands of times faster than the original IBM
PC,
yet the response time today is similar to the response time running Dos
or Windows 3.1
on the early desktop PC's. The culprit is the high overhead of Windows
Operating Systems.

The question really is: If IBM had made a version of MVS which ran on
the Intel x86
architecture, could a modern fast desktop PC handle the load of an IBM
mainframe.
Perhaps it would be a question of throughput.
I wrote a benchmark program in C (which I could email you if you're
interested),
which writes 10 Million 100 byte records to the hard disk, and then
reads them back.
On my Intel Core 2 Duo E6700, writing 10 Million 100 byte records only
took the computer
2 and a half seconds to do, and only a half second to read the 10
Million 100 byte records.
So, potentially, the throughput on desktop PCs is there.

Bill

[Adam Thornton]

Post by Sternbach, William
I wrote a benchmark program in C (which I could email you if you're
interested),
which writes 10 Million 100 byte records to the hard disk, and then
reads them back.
On my Intel Core 2 Duo E6700, writing 10 Million 100 byte records only
took the computer
2 and a half seconds to do, and only a half second to read the 10
Million 100 byte records.
So, potentially, the throughput on desktop PCs is there.

Did the PC actually *write* to disk?

That's only a gigabyte of data. You probably have that much RAM free
on a Core 2 Duo. Memory's pretty fast.

Did you explicitly flush the disk cache after the operation?

Adam

[Non-text portions of this message have been removed]

[Josef Sipek]

Post by Adam Thornton

Post by Sternbach, William
I wrote a benchmark program in C (which I could email you if you're
interested),
which writes 10 Million 100 byte records to the hard disk, and then
reads them back.
On my Intel Core 2 Duo E6700, writing 10 Million 100 byte records only
took the computer
2 and a half seconds to do, and only a half second to read the 10
Million 100 byte records.
So, potentially, the throughput on desktop PCs is there.

Did the PC actually *write* to disk?
That's only a gigabyte of data. You probably have that much RAM free
on a Core 2 Duo. Memory's pretty fast.
Did you explicitly flush the disk cache after the operation?

Also, the filesystem may be adding additional overheads. Ideally, you'd want
to write to something like a raw device (on Unix/Linux):

time dd if=/dev/zero of=/dev/FOO bs=100 count=10000000 oflag=sync

Where /dev/FOO is a physical device. This eliminates the filesystem layer.
There's still the block device layer, but it should do very little. At that
point you have the hard disk's own cache (typically 2-16MB). You can turn
that off as well (smartctl on Linux) but it really depends on how crazy you
want to go.

After you have written, it is very possible that the data is cached. The
simplest way to purge the cache on any system is to reboot :)

You can read those records with:

time dd if=/dev/FOO of=/dev/null bs=100 count=10000000

Just my 2 cents.

Josef "Jeff" Sipek.

[wgs77]

Post by Adam Thornton
Did you explicitly flush the disk cache after the operation?
Adam
[Non-text portions of this message have been removed]

Hi Adam,

Yes, it actually did write to disk. After the benchmark finished
running, I viewed the 100 MB file on the local file system. It
existed and contained the 10 Million 100 Byte rows (as viewed via my
SPFPC editor). If you are interested, let me know, and I could email
you the program source and Windows executable.

Bill

[Adam Thornton]

Post by wgs77

Post by Adam Thornton
Did you explicitly flush the disk cache after the operation?
Adam
[Non-text portions of this message have been removed]

Hi Adam,
Yes, it actually did write to disk. After the benchmark finished
running, I viewed the 100 MB file on the local file system. It
existed and contained the 10 Million 100 Byte rows (as viewed via my
SPFPC editor). If you are interested, let me know, and I could email
you the program source and Windows executable.

Well, right, it got written, but I'm wondering if the write is
asynchronous. That is, the program will return from your write()
call before the data is actually on the platters, and the OS will
flush it to disk at its leisure. Any subsequent process will still
show you the contents of the file, whether they're in RAM or on the
disk, and within, I'd hope, 30 seconds at the outside, you would have
the files actually resident on the disk. That still leaves a lot of
time for the OS to be writing the file that you're not timing.

My question is really, "did you flush the cache after the write
operation while still doing the timing"? (I don't know how you do
that in Windows; it's sync on unix-like systems)

Adam

[Non-text portions of this message have been removed]

[Dave Morton]

High overhead and inefficiency are trademarks of Windows, and
probably other PC operating systems.

I seriously doubt that the 10 million 100 byte records were written
to your PC disk in 2.5 seconds. They probably went to memory in a
disk cache area.

I fully realize that PCs are FAR more powerful than they were in,
say 1985, and still more powerful than in 2000, etc. The
improvements in memory size, CPU power, and I/O speed have been
dramatic and incredible. But I was talking about overall power and
total thruput, etc, not the speed of 1 or 2 benchmarks.

Since your PC does not and CAN NOT run anything remotely close to a
3090's workload, your PC is weaker than a 3090, and the 3090 is FAR
more powerful than your PC!! The 3090 is a 747; your PC is a Piper
Cub.

Why do you cite isolated benchmarks, then leap to the conclusion
that your PC is more powerful than a 3090?? People have been saying
such nonsense since the 1980s. Remember the train analogy and use
logic, as in "comparing apples to apples". A car cannot transport 50
freight cars worth of goods or 300 people(??) with meals across the
country. And your 2007-PC cannot do what a 3090 could do.

There are reasons mainframes are and were expensive, and total
thruput and workload capacity is one of them.

See the document $OSTL18 in the files section.

Dave M

Post by Sternbach, William
Hi,
I believe the reason we don't have 700 users connecting to our

desktop

Post by Sternbach, William
PC
is more a function of the high overhead of modern Windows Operating
Systems.
Desktop PC's today are thousands of times faster than the original IBM
PC,
yet the response time today is similar to the response time

running Dos

Post by Sternbach, William
or Windows 3.1
on the early desktop PC's. The culprit is the high overhead of Windows
Operating Systems.
The question really is: If IBM had made a version of MVS which ran on
the Intel x86
architecture, could a modern fast desktop PC handle the load of an IBM
mainframe.
Perhaps it would be a question of throughput.
I wrote a benchmark program in C (which I could email you if you're
interested),
which writes 10 Million 100 byte records to the hard disk, and then
reads them back.
On my Intel Core 2 Duo E6700, writing 10 Million 100 byte records only
took the computer
2 and a half seconds to do, and only a half second to read the 10
Million 100 byte records.
So, potentially, the throughput on desktop PCs is there.
Bill

[Gerhard Postpischil]

Post by Dave Morton
High overhead and inefficiency are trademarks of Windows, and
probably other PC operating systems.

I have several applications that open and are ready to use while
I'm still clicking under OS/2, that take many, many seconds to
initialize under Windows. While working, the difference is not
as great, but Windows has times when it goes off to play with
itself, and your applications don't get any work done. I don't
know of any other system that approaches Windows sloth by even
an order of magnitude.

Post by Dave Morton
There are reasons mainframes are and were expensive, and total
thruput and workload capacity is one of them.

Don't forget to add reliability and security, to add one more <G>



Gerhard Postpischil
Bradford, VT

new e-mail address: gerhardp (at) charter (dot) net

[Mike Schwab]

An update: ISAM support moved to a separate file with z/OS 1.6 (2005)
and no longer included in z/OS 1.8 orders (2006). (date / version
approximate).

Low address protection - ROM 1975? / EPROM 1992?
Time slicing - perhaps Win 2000 - 2000
LPARs - Virtualization 2006
Sysplex - Clusters 2003?
Stand alone utilities - bootable repair CDs 2000+
VTOC - FAT table and Directory Structure.

On 6/16/07, Dave Morton <marspyrs-***@public.gmane.org> wrote:
<deleted>

Post by Dave Morton
See the document $OSTL18 in the files section.
Dave M

[peter_flass]

--- In hercules-390-***@public.gmane.org, "Sternbach, William " > I wrote a
benchmark program in C (which I could email you if you're

Post by Sternbach, William
interested),
which writes 10 Million 100 byte records to the hard disk, and then
reads them back.
On my Intel Core 2 Duo E6700, writing 10 Million 100 byte records only
took the computer
2 and a half seconds to do, and only a half second to read the 10
Million 100 byte records.

Probably not the best choice of benchmark. Writing, then reading a
sequential file will be heavily skewed by the file system cache. Of
course, it depends on what you're trying to measure, but let's look at
the 700 TSO users. How about a script that edits 700 program files
and compiles them, perhaps repeatedly.

[Dave Wade]

Trying to benchmark the editing and compiling a file by 700 simultaneous
users "interesting". The "obvious", and possibly most realistic way to do
this would be with 700 x TN3270 sessions running scripts with "random"
timers in to simulate the real workload. However as others have pointed out
PC software is a bit top heavy compared with mainframe software. I think my
PC would start to falter if I tried running 70 scripted TN3270 sessions,
never mind 700. However lets assume for the sake of argument it can simulate
70 sessions, then I would still need 10 client PCs to drive the "Mainframe".
I don't even have that in my test lab at work, never mind at home...

.. which is why I have never tried this:-)





Dave.





-----Original Message-----
From: hercules-390-***@public.gmane.org [mailto:hercules-390-***@public.gmane.org] On
Behalf Of peter_flass
Sent: 16 June 2007 19:50
To: hercules-390-***@public.gmane.org
Subject: [hercules-390] Re: IBM 3090 Model 600J compared with modern desktop
PC's.



--- In hercules-390@ <mailto:hercules-390%40yahoogroups.com>
yahoogroups.com, "Sternbach, William " > I wrote a
benchmark program in C (which I could email you if you're

Post by Sternbach, William
interested),
which writes 10 Million 100 byte records to the hard disk, and then
reads them back.
On my Intel Core 2 Duo E6700, writing 10 Million 100 byte records only
took the computer
2 and a half seconds to do, and only a half second to read the 10
Million 100 byte records.

Probably not the best choice of benchmark. Writing, then reading a
sequential file will be heavily skewed by the file system cache. Of
course, it depends on what you're trying to measure, but let's look at
the 700 TSO users. How about a script that edits 700 program files
and compiles them, perhaps repeatedly.





[Non-text portions of this message have been removed]