Some time ago, I ran a benchmark on some aspects of a real-world web application based on gst and iliad. You can see the results here. Some months have passed since then, both Iliad and GNU Smalltalk have moved on. And you can see it!

I had to replace a view message names and categories in the application code to conform to the new API, but apart from that it's the same benchmark on same hardware, slightly updated linux, but both gst and iliad fresh from git master branches:

  n |    dt 1    |    dt 2
    | THEN | NOW | THEN | NOW
 ---+------+-----+------+-----
 10 |    7 |   6 |   19 |  18
 20 |   14 |  12 |   40 |  40
 30 |   22 |  18 |   64 |  59
 40 |   29 |  24 |   82 |  85
 50 |   38 |  31 |  110 | 112
 60 |   45 |  38 |  132 | 131
 70 |   54 |  45 |  160 | 162
 80 |   62 |  51 |  190 | 193 * timeout
 90 |   73 |  58 |  205 | 193 * timeout
100 |   82 |  66 |  230 | 193 * timeout

So we have a visible speedup for the "n simultaneous logins", while the "n complete tests" benchmark stayed more or less the same.

Some really good news can be found with memory consumption: The complete run for n=80, which used 193 MB RSIZE in July, now peaks at only 117 MB, which is a most impressive improvement. Wish I could lose weight as quickly...

NB: The ruby mechanize script weighs in at 250MB RSIZE for this run.

So once again: Kudos to the developers. Great work, friends.