oki, seems to be resolved now. Will perform more tests later today (off to a children party now with my daughter...)
seems that OWB-533 and the JsfPlugin query in the ELResolver were to blame. Performance is now as fast as it never was with 10 parallel threads... LieGrue, strub --- On Fri, 2/25/11, Mark Struberg <[email protected]> wrote: > From: Mark Struberg <[email protected]> > Subject: Re: heavy performance decrease in 1.1.0-SNAPSHOT > To: [email protected] > Date: Friday, February 25, 2011, 9:06 AM > Hi David! > > My tests are not so extreme, but also show a 30% increase > of performance in 1.1.0-SNAPSHOT. > > Of course, this is not yet the same as my real world > application. > I deployed my app to a local tomcat and fired up jmeter > with 20 parallel threads. > Then I simply just changed the owb jars to the old versions > and got a significant performance benefit over our latest > snapshot. > > I did run jmeter a few times before taking the measured > values, so there is definitely no warmup problem. > > A real world app (with MyFaces-2.0.4) additionally uses the > following features: > > * EL integration > * openwebbeans-web > * openwebbeans-jsf > > So I will first try to clone my test and access the values > via EL. > > more to come ... > > LieGrue, > strub > > --- On Fri, 2/25/11, David Blevins <[email protected]> > wrote: > > > From: David Blevins <[email protected]> > > Subject: Re: heavy performance decrease in > 1.1.0-SNAPSHOT > > To: [email protected] > > Date: Friday, February 25, 2011, 5:39 AM > > Spent a few hours poking at > > performance numbers with Grinder and a simple test > case > > based on Marks changes to the interceptor pref > test. > > Seeing 2-3x increase in performance. > > > > Mark, I'm guessing your performance decrease is > thread > > related. Too many. Try dramatically decreasing > > the threads allowed to execute at once. Seems the > > magical number for me was 8 (2-4 per core). > Compiled > > some stats testing behavior against 1.0.x and 1.1.x at > 1, 2, > > 4, 8, 12 ,25 ,50, 100, 500, and 200 threads: > > http://people.apache.org/~dblevins/owb-perf/index.html > > > > > > -David > > > > > > On Feb 24, 2011, at 3:38 PM, David Blevins wrote: > > > > > Hope attachments work > > > > > > <PastedGraphic-1.png> > > > > > > On Feb 24, 2011, at 1:53 PM, Mark Struberg > wrote: > > > > > >> oh that looks really neat. > > >> Will check it out. > > >> > > >> LieGrue, > > >> strub > > >> > > >> --- On Thu, 2/24/11, David Blevins <[email protected]> > > wrote: > > >> > > >>> From: David Blevins <[email protected]> > > >>> Subject: Re: heavy performance decrease > in > > 1.1.0-SNAPSHOT > > >>> To: [email protected] > > >>> Date: Thursday, February 24, 2011, 9:47 > PM > > >>> > > >>> On Feb 24, 2011, at 1:00 PM, Mark > Struberg > > wrote: > > >>> > > >>>> Will do a few parallel unit tests > > tonight. > > >>> > > >>> Side note, I like this tool for parallel > > execution stats. > > >>> > > >>> http://grinder.sourceforge.net > > >>> > > >>> There docs don't have any screenshots, > but > > basically it > > >>> looks like this: > > >>> > > >>> http://people.apache.org/~dblevins/ejbd-client-performance.png > > >>> > > >>> You cook up your own test as you > normally > > would, then it > > >>> will execute it in parallel however you > like > > and take > > >>> samples of each "client" thread and > process > > and spit out > > >>> updated stats. > > >>> > > >>> > > >>> -David > > >>> > > >>>> > > >>>> LieGrue, > > >>>> strub > > >>>> > > >>>> --- On Thu, 2/24/11, David Blevins > <[email protected]> > > >>> wrote: > > >>>> > > >>>>> From: David Blevins <[email protected]> > > >>>>> Subject: Re: heavy performance > > decrease in > > >>> 1.1.0-SNAPSHOT > > >>>>> To: [email protected] > > >>>>> Date: Thursday, February 24, > 2011, > > 6:02 PM > > >>>>> > > >>>>> On Feb 24, 2011, at 9:10 AM, > Mark > > Struberg wrote: > > >>>>> > > >>>>>> oh yes, and I experience > some > > 'stuttering'. > > >>> Means the > > >>>>> log scrolls, then it freezes for > a > > second, logs > > >>> again, > > >>>>> freezes, etc... > > >>>>> > > >>>>> You mean log as in log > > >>> files? Definitely > > >>>>> disable logging when doing any > > performance testing > > >>> or you're > > >>>>> really just measuring your > disk. > > Improved > > >>> logging > > >>>>> could easily explain that sharp > > decrease. > > >>>>> > > >>>>> > > >>>>> -David > > >>>>> > > >>>>> > > >>>>>> > > >>>>>> --- On Thu, 2/24/11, David > Blevins > > <[email protected]> > > >>>>> wrote: > > >>>>>> > > >>>>>>> From: David Blevins > <[email protected]> > > >>>>>>> Subject: Re: heavy > performance > > decrease > > >>> in > > >>>>> 1.1.0-SNAPSHOT > > >>>>>>> To: [email protected] > > >>>>>>> Date: Thursday, February > 24, > > 2011, 4:50 > > >>> PM > > >>>>>>> > > >>>>>>> On Feb 24, 2011, at 8:32 > AM, > > Mark Struberg > > >>> wrote: > > >>>>>>> > > >>>>>>>> hi folks! > > >>>>>>>> > > >>>>>>>> Today I did run a > few > > jmeter tests and > > >>> it > > >>>>> seems that > > >>>>>>> we have some serious > bottle > > neck in our > > >>> code > > >>>>> currently. The > > >>>>>>> app just doesn't really > scale > > well anymore > > >>> and > > >>>>> performance > > >>>>>>> is down to 20% compared > to > > 1.0.0 ... > > >>>>>>> > > >>>>>>> What aspects were you > > measuring and how? > > >>>>>>> > > >>>>>>> > > >>>>>>> -David > > >>>>>>> > > >>>>>>> > > >>>>>> > > >>>>>> > > >>>>>> > > >>>>>> > > >>>>> > > >>>>> > > >>>> > > >>>> > > >>>> > > >>>> > > >>> > > >>> > > >> > > >> > > >> > > >> > > > > > > > > > > >
