Win32 VM, UI Performance & Morphic

Dean_Swan at Mitel.COM Dean_Swan at Mitel.COM
Mon Sep 13 15:55:37 UTC 1999



From:  Dean Swan at MITEL on 09/13/99 11:55 AM

Andreas,

     Your post last week regarding the Win32 VM timing problem
prompted me to do some more investigating this weekend.  The following
code:

     MessageTally time:
     [
       1 to: 1000 do:
       [ :x | Sensor mousePoint ]
     ].

returned results on 4 consecutive runs of:

     Casio E-105, 2.5 img, 2.4c VM       43485, 23890, 24575, 33320

     PII/450, NT Ws4.0 SP 4              5028, 5028, 5027, 5028
       2.5 img, 2.5 VM

Interestingly, I ran this code on my wife's G3/333 iMac with
 2.4c img & VM, but did '1 to: 1000000 do:' instead of '1 to: 1000 do:'
and it returned 7263 !!!  That's about 700 times faster than the PII/450,
and about 3000 to 6000 times faster than the E-105.  These differences
are certainly not due to raw CPU performance alone.  A G3/333 simply isn't
that much faster than a Mips III/131 or a PII/450.


     I think this would have a lot to do with the wide performance
differences that people have mentioned over the last few weeks between
Windoze and Mac (i.e. Morphic is fine on a 120 MHz PowerMac, but too slow
on a 133 MHz Pentium).  This also might explain why Morphic is painfully
slow on the E-105.  It might be worth looking into further?


I also tried this small change on both the E-105 and PII/450:

     Time millisecondsToRun:
     [
       1 to: 1000 do:
       [ :x | Sensor mousePoint ]
     ].

returned results on 4 consecutive runs of:

     Casio E-105, 2.5 img, 2.4c VM       49995, 49985, 50005, 50000

     PII/450, NT Ws4.0 SP 4              5017, 5017, 5027, 5018
       2.5 img, 2.5 VM

The E-105 results strike me as a little bit surprising since
'MessageTally time:' simply calls 'Time millisecondsToRun:'.  I'm
guessing that method cacheing has something to do with this?  The
'MessageTally' version will frequently give results that are as
high as the 'Time' version, but I haven't seen the 'Time' version report
the 50% quicker times at all.  Very curious.


                              -Dean Swan
                              dean_swan at mitel.com





More information about the Squeak-dev mailing list