I was seeing this "quantization" effect earlier when I was running the
benchmarks -- I would see values of 16,666,666 pop up occasionally on a
PII/300 or 25,000,000 on a PII/400. I increased the values in
#tinyBenchmarks to get stable realistic answers. In update 1437 to 2.5
Andreas Raab changed #tinyBenchmark so it automatically compensates by
increasing the values enough to make the benchmarks run for at least 1
second each on any machine. Try getting the last set of 2.5 updates from
the server and see what the benchmarks say.

-- Dwight

Dean_Swan at Mitel.COM wrote:
> 
> Andreas,
> 
>      I just tried an intersting benchmark experiment.  On my PII/450 at work,
> running WinNT/SP4, Squeak 2.5 (through 1432), 2.5 VM (build 6):
> 
> case 1)
>     Open applications under NT -   Psion EPOC32 R5 Emulator,
>                      Win 3.1 Terminal Program,
>                      Premia CodeWrite 5.1b,
>                     Internet Explorer 4.01
> 
>      0 tinyBenchmarks.        25000000 bytecodes/s, 1208527 sends/s
> 
> case 2)
>     Open applications under NT - Win 3.1 Terminal Program,
>                      Premia CodeWrite 5.1b,
>                     Internet Explorer 4.01
> 
>      0 tinyBenchmarks.        16666666 bytecodes/s, 1208527 sends/s
> 
> I'm not entirely sure what this means, other than that the PII/450 is capable of
> performing at the same level as a G3/333 iMac.  The odd thing is that having the
> Psion emulator running seems to make Squeak run faster!
> 
> I don't know if this suggests any VM issues or not, but I thought I'd share the
> raw data point anyway.
> 
>                               -Dean Swan
>                               dean_swan at mitel.com