On Wed, 8 Dec 2004 09:10:02 -0500, Shapira, Yoav <[EMAIL PROTECTED]> wrote: > > When a "spike" appears "for a long time" it's not a spike: it's the > steady state.
Yes, it's the steady state. I just aim at the gc figure, the spike. Maybe this word don't discribes the figure exactly. I am sorry for my English words. > > Please don't jump to determine it's a memory leak just because more > memory is used. I think it is a obvious memory leak because the number of request threads doesn't increase and the app hits just is normal when the inflexion of gc figure appreas. The question only is caused by my app, or tomcat. > > So you're setting the heap size to a constant 1250MB. How can there be > big jumps all over the place? It should been given max memory as possible to avoid JVM increases heap progressively when the server machine only runs tomcat. I think all heavy load app should set two parameter equal. > I bet your app is more likely than Tomcat to have a leak ;) Simple JSPs > can cause memory "spikes," as can simple servlets. DBCP and threads do > NOT necessarily cause increased memory usage. It is probably that my app has a leak. I agree. But every JSP has been visited during four days of system in order. Why to appear spike at fifth day? I have 2~3 hundreds JSPs that almost be requested everyday. Does one JSP has memory leak at a special situation? Or one JSP running at a special situation causes tomcat leak? > > Get a test system, put your app on it, run it with a profiler, and > simulate load using a test tool of your choice, e.g. JMeter, ab, wget, > grinder, whatever. Then see where memory is retained during your > "spikes." > Thanks for the advice. and thanks for Dale's suggustion. I will try to profile heap through -Xrunhprof and other tools. Li Zhenxing --------------------------------------------------------------------- To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED]
