olson@modular.UUCP (Jon Olson) (05/29/88)
I decided to test out some of the theories on the NET about
Vax's being better at handling alot of jobs and high user loads.
The benchmarks below show the elapsed time required for 8
simultaneous optimized compiles of a 3783 line C program on
each machine, with no other jobs currently running. Here are
some stats on the machines with the following configurations:
1) Vax 11/780, 8Mb memory, VMS V4.6, VAX-C V2.4
2) MicroVax, 9Mb memory, Ultrix V2.0, Unix PCC compiler
3) Sun 3/60, 20Mhz 68020, 8 Mb memory, SUNOS V3.4
-Time (secs)- -Relative Performance-
Machine CPU Elapsed CPU/Elapsed CPU Elapsed
------- ----- ------- ----------- ----- -------
11/780 1206 1728 69.8% 1.0 1.0
MicroVax 1206 1370 88.2% 1.0 1.3
Sun 3/60 498 551 90.4% 2.4 3.1
The CPU time in the above table is the sum of all jobs; the elapsed
time is the time for the last job to finish. All jobs were started
simultaneously and had exclusive use of the machine during execution.
Note from the above table that the Sun 3/60, (yes running a 68020)
had the highest ratio of (CPU time / elapsed time) and the worst
performer was, as I would have guessed, the Vax 11/780 running VMS.
Also interesting was that the 11/780 and MicroVax had identical
CPU times even though running different compilers on different systems.
The VMS overhead, however, makes the VAX-C compiler much slower
than the portable C compiler on Ultrix V2.0. Seems to me that the
system that does poorly under load is Vax's running VMS!
--
Jon Olson, Modular Mining Systems
USENET: {ihnp4,allegra,cmcl2,hao!noao}!arizona!modular!olson
INTERNET: modular!olson@arizona.edudaveb@llama.rtech.UUCP (It takes a clear mind to make it) (06/02/88)
In article <603@modular.UUCP> olson@modular.UUCP (Jon Olson) writes: > >I decided to test out some of the theories on the NET about >Vax's being better at handling alot of jobs and high user loads. >The benchmarks below show the elapsed time required for 8 >simultaneous optimized compiles of a 3783 line C program on >each machine, with no other jobs currently running. Here are >some stats on the machines with the following configurations: > > 1) Vax 11/780, 8Mb memory, VMS V4.6, VAX-C V2.4 > 2) MicroVax, 9Mb memory, Ultrix V2.0, Unix PCC compiler > 3) Sun 3/60, 20Mhz 68020, 8 Mb memory, SUNOS V3.4 A more interesting version of this would run 1 2 4 6 8 12 16 20 users or so, allowing a graph of results by load. Any volunteers? -dB {amdahl, cpsc6a, mtxinu, sun, hoptoad}!rtech!daveb daveb@rtech.uucp
sjr@mimsy.UUCP (Stephen J. Roznowski) (06/13/88)
In article <2133@rtech.UUCP> daveb@rtech.UUCP (It takes a clear mind to make it) writes: >In article <603@modular.UUCP> olson@modular.UUCP (Jon Olson) writes: > >A more interesting version of this would run 1 2 4 6 8 12 16 20 users or >so, allowing a graph of results by load. Any volunteers? > >-dB This is easily done by using MUSBUS which was posted in comp.sources. [MUSBUS stands for Montash University Suite for Benchmarking Unix Systems, and allows you to specify jobs for multiple users to execute and does it for any number of users -- See MUSBUS for more details]. Stephen -- Stephen J. Roznowski sjr@mimsy.umd.edu