Peter Lin wrote: > I just ran some benchmarks with 4.1.6 and noticed a > couple of things. > > 1. 4.1.6 is showing less of an impact once the > connector reaches the max processors. With 4.0.3, I > was able to trigger a out of memory error by letting > JMeter continue for a couple minutes after tomcat > reaches the max processors and is denying connections. > I discovered this by accident when I was running > benchmarks with 120 con current connection hitting > pages with lots of JSTL tags. the max processor was > set to 75. > > 2. performance appears to be more consistent memory > and cpu wise. Under moderate and heavy load, the cpu > usage varies more with 4.0.3 and it takes longer for > 4.0.3 to return to normal memory usage. This was done > with just task monitor and not optimizeIt or JProbe > since I can't get JProbe to work correctly with > jasper2. > > 3. tag pooling does show 5-10% improvement over > jasper2 w/o tag pooling.
I haven't profiled Jasper 2 for a long time, so I'm not sure how complete the optimizations are (it should be quite good at this point though, since Costin probably did). I'd say tag pooling may help more on pages which use lots of simple tags like some i18n tag. It still obviously helps reduce GC, which never hurts. Did you set the "development" flag to false while testing ? I think you're supposed to use that in production systems now (I suppose we need a doc page on Jasper 2, as there are quite a few new configuration parameters which are actually useful now). > Once I figure out how to get JProbe to work with > jasper2, i will run more tests. Remy -- To unsubscribe, e-mail: <mailto:[EMAIL PROTECTED]> For additional commands, e-mail: <mailto:[EMAIL PROTECTED]>
