-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1 Jorge Martín,
Jorge Martín Cuervo wrote: > It's a realy cool question, i've a complex webapp with a lot of > differents diciplines: > > - xstl > - xsl-fo > - very large objects in session These three are going to kill you. XSLT uses a ton of memory during processing, depending on how your templates are written. I would imagine that the processor could really work like a SAX pipeline should and be fairly lean and mean, but as soon as you start running xpath expressions like '/root//something' you can require that entire XML documents be cached in memory; not just the text itself, but tons and tons of associated objects. > The tests with jmeter using 10 concurrents users don't cause an memory > exception (i use 256M for heap) That's good. > there is some guide to view what we have to check to see what's > happening behind the scenes? I would turn on verbose GC in your VM and watch the memory usage during your load tests. Run it for a few hours with a load of 10 users, and then take the GC output and graph it. If you get a nice curve that looks like this: /\/\/\/\/\/\/\/\/\/\/\ then you are probably okay in terms of application memory leaks (meaning that you have few or none). If it goes steadily upward, you might want to look into what those objects are. If your load test doesn't run out of memory, you'll have a memory baseline for 10 users. Ramp that up to 15 or 20 users and repeat the entire process. Make sure that the memory curve still looks like /\/\/\/\/\ and then you'll have a measurement for 15 (or 20) users. Try this again with another step (maybe 25 or 30) and check the numbers. - From those 3, you ought to be able to predict how much memory you need for a certain number of users. Determine the peak load that you want to support, and set the heap to that size. Then, sit back and watch. It might be good to know the number of users logged-into your system at all times. If you get an OOM, you might be able to see that you got more users than you predicted (maybe you planned for 50, and had 58 instead). In that case, you are free to increase the heap size accordingly. If there were only 10 users on during the OOM, then you either have a special case where a ton of memory is required (very possible) or a problem like a memory leak. Hope that helps, - -chris -----BEGIN PGP SIGNATURE----- Version: GnuPG v1.4.5 (MingW32) Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org iD8DBQFFS0c/9CaO5/Lv0PARAjcsAJ9L/9faaE8W/b2jIFjPO+6oQpgbPgCfc9zK g/Hz30E19Et9lKUhLHmxXQA= =xzbR -----END PGP SIGNATURE----- --------------------------------------------------------------------- To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED]
