Lies, Damned Lies and ...

The following measurements are derived from benchmark code published on the Smalltalk Industry Council site, under 'Benchmarks'.  Code is available here only so that anyone wishing to reproduce the measurements may do so readily in the various images.

Tests were conducted on a 166MMX Pentium with 40MB RAM, under Windows/95 OSR2.  Each test was run three times.
Test Smalltalk Express 2.0.4 Visual Smalltalk 3.0.1 Squeak 1.20 Squeak 1.30 Squeak 1.30J Dolphin/98 rel1 VAST 4.02 Smalltalk MT 1.5b3 Java 1.2b2
alloc 3.9 1.1 3.0 3.1 3.9 1.9 2.5 15.7 846.8(1)
arrayWrite 8.4 2.4 11.9 11.8 9.9 9.1 4.7 1.7 0.5
dictionaryWrite 5.9 0.5 3.6 3.8 3.4 1.6 0.5 1.0 6.7
floatMath 20.7 7.1 19.3 20.5 24.5 14.1 17.8 66.8 2.6
integerMath 20.5 0.9 7.1 7.0 5.9 2.3 1.8 1.0 3.2
collectionIterate 21.8 5.9 42.1 42.2 34.7 24.5 4.0 3.2 61.6
collectionWrite 3.5 1.0 6.3 6.4 5.4 4.0 2.0 2.9 4.1
stringCompare 2.3 0.9 301.2 241.4 215.7 0.6 1.2 3.6 0.9
hanoi 4.7 0.9 12.6 12.9 7.9 8.2 1.8 1.0 4.0
  Source --> perfstex perfstex perfsqek perfsqek perfsqek perfdolp perfvast perfstmt perfjava hanoi

Notes:

  1. Extrapolated from 1000 iterations, rather than the 100,000 used in other images.
  2. Some comments from ObjectConnect on the Smalltalk/MT.