First off, note that the maximum resolution available on OS X is microsecond level.
This means that any timings of any intervals of the order of 1us (or less) will be reporting whether or not you caught a clock edge, not the actual timing interval. You can probably model this using a uniform probability distribution (or some slight modification thereof) - but I'm not sure how much that's going to help. Ben On Mon, Mar 5, 2012 at 2:56 AM, Mark Roos <mr...@roos.com> wrote: > I have been working on execution time for my Smalltalk implementation and > am having trouble getting repeatable times. I am using a script from a new > launch of java with all apps turned off. I run the target code 10 times to > let > Hotspot do its thing. What I see are times of great performance and times > that are 30% slower after they stabilize. Since I am trying fine grained > changes > in method lookups etc it makes it difficult to determine if what I try helps > or hurts. > > Any suggestions on how to do a better job? I am using a core 2 duo with > snow leopard. > > thanks > mark > > > _______________________________________________ > mlvm-dev mailing list > mlvm-dev@openjdk.java.net > http://mail.openjdk.java.net/mailman/listinfo/mlvm-dev > _______________________________________________ mlvm-dev mailing list mlvm-dev@openjdk.java.net http://mail.openjdk.java.net/mailman/listinfo/mlvm-dev
