[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: Getting task elapsed ti

                       Subject:                               Time:9:46 AM
  OFFICE MEMO          RE> Getting task elapsed time          Date:11/26/1993
> The context is that I need to determine the time taken to issue a pen
> with an accuracy of up to 10 milliseconds, if possible.  Although the
> get-internal-run-time variable says that (get-internal-run-time) is accurate
> to 1000 milliseconds, would this accuracy still hold if GC is ... on?

The time will be "accurate" but it will include any time spent garbage
collecting.  During this time the user/subject may be prevented from concluding
their response to whatever you're timing.  An easy fix that works in many cases
is to force a garbage collection (by calling gc) just before each trial.  In
many cases, this sharply reduces the probability that garbage collection will
happen during the course of a timed trial.

If garbage collection is still happening, and you really need accurate
performance time estimates, you could count the number of garbage collections
that happen for each subject, and try removing the effect due to garbage
collection using regression/ANOVA.  I'd be careful that the "waiting for gc to
finish" task doesn't somehow interact with the way the subjects perform your
primary task, though.  But this too is a testable hypothesis given enough data.