Filip Hanik - Dev Lists wrote: > Mark Thomas wrote: >> I was a little surprised that ThreadLocal seems to be slower. I still >> wonder if I got the test wrong for that but I can't see anything. >> > If it was me doing the date, and making it all threadlocal, with no > global variables at all > And this will run faster, even though initially creating the thread > local map incurs overhead, it doesn't do unnecessary object creation or > assignment. And there is no need, since we are local at all times.
Thanks for the review. I think I was having a "Can't see the woods for the trees" moment. I've made the changes you suggested and lo and behold the ThreadLocal is quicker. > testAccessLogGetDate: 16 threads and 10000000 iterations using Syncs > took 7461ms > testAccessLogGetDate: 16 threads and 10000000 iterations using > ThreadLocals took 6433ms Out of interest, how many cores are you running? With 4 cores I get a bigger difference: org.apache.catalina.valves.Benchmarks$TimeDateElementBenchmarkTest_Sync: 16 threads and 10000000 iterations using Syncs took 14359ms org.apache.catalina.valves.Benchmarks$TimeDateElementBenchmarkTest_Local: 16 threads and 10000000 iterations using ThreadLocals took 1532ms Per request this is still next to nothing. I did think about not bothering to port the change at all. However, given the possibility for contention with the various syncs on 'this' in the AccessLogValve, I will port it as it will be quicker to port it than it would be to write a better, more representative benchmark for the AccessLogValve. Mark --------------------------------------------------------------------- To unsubscribe, e-mail: dev-unsubscr...@tomcat.apache.org For additional commands, e-mail: dev-h...@tomcat.apache.org
