On Sat, 08 Jul 2006, Charles O Nutter defenestrated me: > > I'm starting to run some profiles to find real hotspot problems. So > here's the first summary of findings. > Profiling: today's code, gem install --force rake (gem cached > locally), up to some time during ri generation (killed after an hour) > This profiling run just examined CPU times for various method calls. > Based on "own time" the amount of time spent in each method, there are > a few standouts. > - ThreadLocal.get is number one again, accounting for 4% of global > execution time. If we modified JRuby such that the current thread's > ThreadContext was passed along the call chain we might be able to > eliminate much of this overhead. ThreadService.getCurrentContext() and > Ruby.getCurrentContext() are ranked #4 and #6 and take another 2% > each, adding up to roughly 8% of total execution time spent retrieving > the current ThreadContext. EvaluationState.getThreadContext adds > another 1%.
We also talked about pushing method calls into ThreadContext itself for execution. That is probably a larger change, but it will kill the need to call getcurrentContext() alot (I think). > - UnsynchronizedStack.peek and ArrayList.size are #2 and #3, certainly > owing to our heavy use of UnsynchronizedStack in the core interpreter. > Not too surprising, but the more we can eliminate the need for them > both the better. They accounted for 6% of execution time. > ArrayList.size was invoked 116 million times during my test run. We also identified two possible Stacks marked for death: RubyClass (for CRef spaghetti) and Scope merging with Frame? > - RubyModule.findImplementer = 6%, If we address overall problem with method cache then I think we can get rid of this. > - ThreadContext.preReflectedMethodInternalCall and callees took up > 12%, which is a bit surprising since it doesn't do all that much. A > large part of that is pushing a new frame onto the frame stack. > - RubyThread.criticalizeOrDieIfKilled alone was 6%, contributing to > pollThreadEvents above I wish we understood the right granularity to check the thread stuff. I am sure we are doing it more than we need, but I cannot see a good injection point. -Tom -- + http://www.tc.umn.edu/~enebo +---- mailto:[EMAIL PROTECTED] ----+ | Thomas E Enebo, Protagonist | "Luck favors the prepared | | | mind." -Louis Pasteur | ------------------------------------------------------------------------- Using Tomcat but need to do more? Need to support web services, security? Get stuff done quickly with pre-integrated technology to make your job easier Download IBM WebSphere Application Server v.1.0.1 based on Apache Geronimo http://sel.as-us.falkag.net/sel?cmd=lnk&kid=120709&bid=263057&dat=121642 _______________________________________________ Jruby-devel mailing list Jruby-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/jruby-devel
