From:  Dean Swan at MITEL on 08/19/99 10:54 AM

0123456789012345678901234567890123456789012345678901234567890123456789

John,

     I ran the Squeak Benchmarks on a Casio E-105 Running Andreas'
WinCE/MIPS 2.4VM with a virgin 2.5 image in the standard UI several
times.  The results are as follows:

          2857142 bytecodes/s
           115752 sends/s

          2857142 bytecodes/s
           144344 sends/s

          2857142 bytecodes/s
           140021 sends/s

          2857142 bytecodes/s
           138606 sends/s

     For comparison, Andreas' Win32 2.4VM with a virgin 2.5 image on
a 450 MHz Pentium II running NT Workstation 4.0 gives:

          16666666 bytecodes/s
           1059173 sends/s

So, the 450 Pentium II is on the order of 8 times faster than the E-105.


     I've also looked into a few other things that Andreas suggested.
I added:

     '               ' printString displayAt: 0 at 0.
     wait printString displayAt: 0 at 0.

to Controller interActivityPause, and PasteUpMorph interCyclePause,
just after 'wait' is calculated in both methods.

     In the standard UI, the value of wait is typically in the
-400 to -700 range, occasionally in the -1000 to -1500 range, and very
rarely up in the -3000 to -4000 range.  In Morphic, the value of wait
is *frequently* in the -20000 to -30000 range, and sometimes in the
-700 to -1500 range.

     Andreas suspected that the interCyclePause: was cause the delays
in Morphic, but from these tests, it's pretty clear that that's not the
problem.  It also indicates that something else is often going on that
takes 20 to 30 seconds during an iteration of 'doOneCycle'.

     I'll continue to track down where the time is going, and post
anything that I discover, but any suggestions would be welcomed.


                              -Dean Swan

P.S. You guys did some serious redecorating in the Morphic classes
in the 2.5 image.  A lot of stuff that was in WorldMorph seems to have
relocated to PasteUpMorph.  What motivated that?