donn@rice.EDU (Donn Baumgartner) (09/17/87)
In response to my claim that the AT (roughly) benches above the vax,
Larry McVoy basically complained that he had seen just the opposite,
albeit "only a little tiny bit slower", and proceeded to give standard numbers
associated with AT machine's disk speeds, cpu speeds, and memory count.
The jist being that an AT running some non-BSD un*x system appeared to be
slower than a vax 750 (probably not running xenix).
To this, Ron Natalie (basically) responds that Larry's comparison is not
a fair one.
------------------------------------------------------------------------
Apples and oranges? Maybe. Let me clarify just a little:
(1) the benchmarks are not comprehensive, and did not include disk bench
marks (I said simple benchmarks, and I meant it);
(2) a standard AT disk has a '28 ms' random access time, it's sequential
read time (for loading, etc.) is typically 2 to 3 times that;
(3) if you spent as much money on an AT disk as you will for a vax disk,
the AT would fly;
(4) an AT with only 1.5Mb of RAM running BSD or any other un*x is a crippled
machine (how much memory did your vax 750 have?) (I should mention that
AT RAM is considerably cheaper than vax RAM);
(5) the 4.3 port to the AT is using the bsd fast filesystem - theoretically
that will make it faster than your average xenix filesystem
(6) the drystone benchmark (a pseudo-typical C program) shows one very
important thing, that not all C compilers are created equal. I have
run that benchmark under DOS (3 compilers), PC/IX (it's compiler), and
a xenix (it's compiler); all the times were different, with pc/ix taking
top honors at '1215' (8MHz AT)... some of the times were slower than the
vax time on the same code (but *different* compiler).
(7) the simple bench we used for comparison ... a shell scipt which spawns
/bin/echo many times in a loop, runs faster on the AT than the vax.
The usefulness of that bench is also questionable.
OK, so what's my jist? Well, first, I only use the vax for comparison because
it has in some sense become a standard to compare against. My point was that
an AT running 4.3 bsd should make a cost effective, reasonably fast, personal
work station. It's not a vax, or a sun, and should not be directly compared
with either (especially on pricing :-). And then there's my bias; why would
I tell you negative things about the machine I'm helping port 4.3 to...
- Donn Baumgartnerallyn@sdcsvax.UUCP (09/17/87)
In article <9338@brl-adm.ARPA>, donn@rice.EDU (Donn Baumgartner) writes: > In response to my claim that the AT (roughly) benches above the vax, > Larry McVoy basically complained that he had seen just the opposite, > albeit "only a little tiny bit slower", and proceeded to give standard numbers > associated with AT machine's disk speeds, cpu speeds, and memory count. > The jist being that an AT running some non-BSD un*x system appeared to be > slower than a vax 750 (probably not running xenix). i've just been doing some (trivial) benchmarks of an IBM PC/AT (and the other PC models), a VAX-11/780 and a SUN-3/110. they might be of slight interest to somebody, so here they are. please no flames about this benchmark. it obviously doesn't test any i/o, context switching, system calls or anything else. just some probably useless measurement of processor speed. all timing was done with either the csh built-in "time" or /bin/time. all machines were in multi-user mode, so the time reported is just the sum of the user and system time, not the elapsed time. operating systems: VAX 4.3 BSD (looks to have an FPA from the benchmarks) SUN SunOS 3.2 PC/AT PC/IX 1.1 (has an 80287, ~8.5 MHz processor speed) PC/XT PC/IX 1.1 (has an 8087) PC PC/IX 1.1 (uses PC/IX software floating point simulation) this (admittedly stupid) benchmark (10,001,000 additions/increments): main() { register x, y; for (y = 0; y < 1000; y++) for (x = 0; x < 10000; x++); } took the following times (seconds user+system, not wall clock): sun-3/110 (chems1.ucsd.edu) 8.5 vax-11/780 (sdcsvax.ucsd.edu) 22.2 ibm pc/at (robin.emu.ucsd.edu) 26.3 ibm pc/xt (sdemu.ucsd.edu) 101.1 the same program with the declaration line changed to: int x, y; (without registers) took: sun-3/110 15.8 vax-11/780 33.2 ibm pc/at 45.3 ibm pc/xt 205.9 this floating point program (1,000,100 floating point add/increments): main() { float x, y; for (y = 0; y < 100; y++) for (x = 0; x < 10000; x++); } took: vax-11/780 20.5 sun-3/110 52.5 ibm pc/at 88.6 ibm pc/xt (sdemu.ucsd.edu) 135.0 (8087) ibm pc (kiwi.emu.ucsd.edu) 5701.7 (software fp) (that is NOT a typo) (that's over 1.5 HOURS) and the same program with variables changed to double from float: vax-11/780 17.2 sun-3/110 65.7 ibm pc/at 95.2 ibm pc/xt 146.2 (8087) ibm pc ? (software fp) (i wasn't about to wait) -- From the virtual mind of Allyn Fratkin allyn@sdcsvax.ucsd.edu or EMU Project {ucbvax, decvax, ihnp4} U.C. San Diego !sdcsvax!allyn
Leisner.Henr@Xerox.COM (marty) (09/17/87)
Donn, FYI, My timings for Dhrystone with Aztec C 3.4 -- [0:811] dry Dhrystone(1.1) time for 50000 passes = 32 This machine benchmarks at 1562 dhrystones/second This is with several 186 optimization switched on. marty GV: leisner.henr NS: martin leisner:henr801c:xerox UUCP: martyl@rocksvax.uucp
allyn@sdcsvax.UUCP (09/17/87)
In article <3891@sdcsvax.UCSD.EDU>, I posted the following benchmark: > main() > { > float x, y; > > for (y = 0; y < 100; y++) > for (x = 0; x < 10000; x++); > } with the following times (in seconds user+system time, vax included for comparison): > vax-11/780 20.5 > sun-3/110 52.5 > > and the same program with variables changed to double from float: > vax-11/780 17.2 > sun-3/110 65.7 david@sun.com (David DiGiacomo) pointed out that i was using software floating point on the sun. so i re-ran the benchmarks for the sun using both the 68881 and the fpa. floats: sun-3/110 6.0 (fpa) vax-11/780 20.5 sun-3/110 22.9 (68881) sun-3/110 52.5 (software fp) doubles: sun-3/110 9.9 (fpa) vax-11/780 17.2 sun-3/110 21.6 (68881) sun-3/110 65.7 (software fp) -- From the virtual mind of Allyn Fratkin allyn@sdcsvax.ucsd.edu or EMU Project {ucbvax, decvax, ihnp4} U.C. San Diego !sdcsvax!allyn
allbery@ncoast.UUCP (09/24/87)
As quoted from <9338@brl-adm.ARPA> by donn@rice.EDU (Donn Baumgartner): +--------------- | (5) the 4.3 port to the AT is using the bsd fast filesystem - theoretically | that will make it faster than your average xenix filesystem +--------------- I think I already nipped this one in the bud; the Altos 886 (80286, and roughly the same controler and definitely the same drives as the AT) can be used with the "BoosterPak", which is a BSD fast file system, or with the standard Xenix filesystem. I've noticed very little difference between the two. -- Brandon S. Allbery, moderator of comp.sources.misc {{harvard,mit-eddie}!necntc,well!hoptoad,sun!mandrill!hal}!ncoast!allbery ARPA: necntc!ncoast!allbery@harvard.harvard.edu Fido: 157/502 MCI: BALLBERY <<ncoast Public Access UNIX: +1 216 781 6201 24hrs. 300/1200/2400 baud>> "Mummy, what's an opinion?"