Cache misses count as CPU time. I've seen people measure parallel speedup by looking at CPU usage. They get linear increase in CPU use to eight threads even though the task saturated memory bandwidth after the second thread.
If you have a way to measure instruction count you can compare increase in instruction count in GC to increase in time in GC. If instructions are constant while time increases you probably have a memory access problem. If instructions increase the structure of references within your data may be forcing the collector to do more work. > However, the application still consumes more and more CPU time. And it > seems to happen in the GC. Apart from that, the application seems to > be just fine. So it seems to be something in our code (or in LablGTK) > that is making the GC spend more and more time. Anyone experienced > this kind of problem? -- Caml-list mailing list. Subscription management and archives: https://sympa-roc.inria.fr/wws/info/caml-list Beginner's list: http://groups.yahoo.com/group/ocaml_beginners Bug reports: http://caml.inria.fr/bin/caml-bugs
