After much work w/ Peter Williams and Con Kolivas, I finally have a decent performing genetic-lib implementation on interbench. It does orders of magnitude better in the burn & compile pieces. What really screws Zaphod and the genetic-lib up is when there is one interactive child running in the background w/ a bunch of threads that are doing while(1) {}. The interactive task does not get classified as such and they miss many of their checkpoints while the cpu hogs run. This latest version has a few bug fixes and detects interactive tasks sooner. The only thing to watch is that with the quicker detection of CPU hogs, a burst of CPU activity on X will get it classified as a CPU hog. Please let me know if you see as such.

I’ve been using interbench a lot lately, and it does an excellent job of catching the “bad children”. I can see almost immediately when some child gets mutated off into the woods and performance tanks. Con did a great job with the benchmark.