Here are some benchmarks using a modified version of the STIC benchmarks
(attached), all run on WinNT4sp6a on a PII/400. I would agree that the
magnitude of speedup indicated by #tinyBenchmarks from 2.7 to 2.8 is not
generally apparent (but this has been true of microbenchmarks on
Smalltalk VMs since the beginning). One must always look at the
percentage of total time the VM is actually spending in primitives --
speedups like that shown by #tinyBenchmarks can only effect the
remaining percentage of time spent outside of primitives. There are some
solid speedups for several of the benchmarks though (and a few that are
actually slightly slower).

-- Dwight

++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
Squeak 2.7#1782 on 2.7 Win32 VM (release build 2 of Jan 8, 2000)

Begin Speed Test
18680677 bytecodes/sec; 1007303 sends/sec
4747ms - simple Hanoi.
991ms - allocation.
3815ms - array write.
1312ms - dictionary write.
7971ms - floating point math.
2825ms - integer math.
14070ms - OrderedCollection iterate.
2323ms - OrderedCollection write.
4026ms - string compare.
End Speed Test

Begin Speed Test
18691588 bytecodes/sec; 1072215 sends/sec
4747ms - simple Hanoi.
1001ms - allocation.
3726ms - array write.
1322ms - dictionary write.
7981ms - floating point math.
2824ms - integer math.
14020ms - OrderedCollection iterate.
2413ms - OrderedCollection write.
4026ms - string compare.
End Speed Test
++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
Squeak 2.8a#2335 on Win32 VM of May 24, 2000

Begin Speed Test
41775456 bytecodes/sec; 1186932 sends/sec
3745ms - simple Hanoi.
931ms - allocation.
2744ms - array write.
1492ms - dictionary write.
8372ms - floating point math.
1942ms - integer math.
10936ms - OrderedCollection iterate.
2464ms - OrderedCollection write.
4046ms - string compare.
End Speed Test

Begin Speed Test
41775456 bytecodes/sec; 1178526 sends/sec
3775ms - simple Hanoi.
922ms - allocation.
2744ms - array write.
1482ms - dictionary write.
8462ms - floating point math.
1943ms - integer math.
10946ms - OrderedCollection iterate.
2463ms - OrderedCollection write.
4076ms - string compare.
End Speed Test
++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++


Stefan Matthias Aust wrote:
> 
> At 11:50 18.06.00 -0500, Dwight Hughes wrote:
> >#tinyBenchmarks is just running #benchmark and #benchFib - neither of
> >these creates LargeIntegers. For my system (a PII/400) I'm getting:
> >
> >42049934 bytecodes/sec; 1146059 sends/sec
> 
> On my PII/400 (actually overclocked to 448 MHz) I get
> 
> '41423948 bytecodes/sec; 1254015 sends/sec'
> 
> For a long time now, I distrust that test, especially since no other
> benchmark I tried could prove Andreas' reported speedup between VM 2.7 and
> 2.8.
> 
> BTW, on a dual process SPARC 400 MHz system, I get about 20 millions
> b/s.  I can't believe that this machine is so much slower.  At least,
> Squeak doesn't feel so much slower (not counting the X display which is -
> as I always work remote - slower)
> 
> I don't understand why we don't create a benchmark which not just believes
> that it is counting bytecodes but actually counting them - it shouldn't be
> too difficult to really calculate them.  Perhaps even the VM can count it -
> I recently saw some profiling primitives in SystemDictionary.
> 
> bye
> --
> Stefan Matthias Aust  //  Bevor wir fallen, fallen wir lieber auf
-------------- next part --------------
A non-text attachment was scrubbed...
Name: Benchmarks-STICTests.cs
Type: application/x-unknown-content-type-editplus.7
Size: 11051 bytes
Desc: not available
Url : http://lists.squeakfoundation.org/pipermail/squeak-dev/attachments/20000618/558f1be0/Benchmarks-STICTests.bin