I don't use CacheManager (since that would defeat the purpose of the test - which is to exercise the entire system for all assets from the pages being requested).
I have logged the GCs (without removing the plugins), and it doesn't do any GCs at all during the 'pause' period. The GC right before the pause takes less than a second (.26...) and the one right after is in the same ball-park of duration. But during the pause, no GCs at all occur. CPU goes to idle on the JMeter box during the pause (as does network and disk I/O, and swap usage). There is a _little_ activity, but I've been attributing that to the Remote Desktop I'm using to get to the box, and the PerfMon graphs that are still collecting data. But compared to before and after the pause the activity level is less than 2% during the 'pause' period, compared to 75-80% during the 'run' period. I will try to get the thread dump - I'm working on some production issues today, so I haven't had time to setup for a thread dump today. -- Robin D. Wilson Sr. Director of Web Development KingsIsle Entertainment, Inc. VOICE: 512-777-1861 www.KingsIsle.com -----Original Message----- From: Philippe Mouawad [mailto:[email protected]] Sent: Thursday, December 15, 2011 2:23 PM To: JMeter Users List Subject: Re: Strange 'pause' activity during testing Do you use CacheManager ? You should remove any plugin and activate GC logs to check it's not GC ? How is CPU on JMeter stack ? Regards Philippe On Thu, Dec 15, 2011 at 9:09 PM, Robin D. Wilson <[email protected]> wrote: > >Maybe then there is a problem with the scanning of the HTML to > >extract the > embedded resources, or maybe one of the embedded resources is a tar-pit. > > If this were the case, I would expect the first sample to show the problem. > The fact that it does 600+ iterations without a problem - and _then_ > stalls seems like it rules out any problem with the returned HTML > (especially since the only difference in the returned result is the > username supplied by JMeter). > > >Do all the expected embedded resources get downloaded? > > Zero errors (even with the pauses there are no errors at all). > > >Are there any unusually long elapsed times for embedded resources? > > I see the 'max request duration' jump up right after the pause - but > it is only 10-11 seconds (not ~35 seconds like the pause). > > >Or large gaps between the parent sample download completion and the > >start > of the first embedded resource? > > Not that I can see... I'll see if I can get more detail on this. > > >That would suggest the page took a while to parse. > > I would assume that because this happens ~600+ iterations into the > test (the first time), that if it was related to parsing the page, I > would see it earlier in the test run cycle. And I wouldn't expect a > parsing problem to repeat on such a consistent basis - without > happening on every sample. > Right > now, if it is a parsing problem, it only happens the ~600th time it > sees the same page, which seems really surprising to me (then it > happens again after another ~600 iterations, etc.). Also, I would > expect the parser to take up some CPU and perhaps even some I/O > cycles, but the PerfMon shows idle during the pause period. > > >You'll need to select the optiion to "save subresults" in order to > >see the > embedded samples. > > ... > > > >BTW, you wrote you were running JMeter 2.4.1 - that does not exist, > >so > perhaps you meant the current version, 2.5.1? > > Sorry, I meant 2.4. I didn't upgrade to 2.5 (and beyond) because of a > previously reported problem where 2.5+ slows down my throughput to > about 60% of what I get on 2.4. > > > --------------------------------------------------------------------- > To unsubscribe, e-mail: [email protected] > For additional commands, e-mail: [email protected] > > -- Cordialement. Philippe Mouawad. --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
