Loreno Oliveira [http://community.jboss.org/people/lorenooliveira] created the discussion "Re: JBoss memory usage pattern" To view the discussion, visit: http://community.jboss.org/message/588863#588863 -------------------------------------------------------------- Yes, that's exactly how I'm doing. I'm watching the server behavior using visualvm. Is there a better approach to monitor an application but JMX? I mean, how could one get data about some application health with a minimum of overhead and a maximum of accuracy? -------------------------------------------------------------- Reply to this message by going to Community [http://community.jboss.org/message/588863#588863] Start a new discussion in Performance Tuning at Community [http://community.jboss.org/choose-container!input.jspa?contentType=1&...]

Loreno Oliveira [http://community.jboss.org/people/lorenooliveira] created the discussion "Re: JBoss memory usage pattern" To view the discussion, visit: http://community.jboss.org/message/589010#589010 -------------------------------------------------------------- Hi Wolf-Dieter, Actually, I started all this observation after reading about gc performance tunning. I'm trying to find evidences that lead me to make some sort of tunning. So far, I didn't find anything in this sense, but I found this behavior. The strange point here is the volume. I mean, the heap usage is growing almost 2GB in 40s with low (or none) workload, and in 8s with a higher workload. Is this normal (in this volume)? In the point of view of the users, we are not getting malfunction reports.However, we should grow our workload in the next weeks... -------------------------------------------------------------- Reply to this message by going to Community [http://community.jboss.org/message/589010#589010] Start a new discussion in Performance Tuning at Community [http://community.jboss.org/choose-container!input.jspa?contentType=1&...]

Loreno Oliveira [http://community.jboss.org/people/lorenooliveira] created the discussion "Re: JBoss memory usage pattern" To view the discussion, visit: http://community.jboss.org/message/589072#589072 -------------------------------------------------------------- Peter, I understand the point that visualvm can be, itself, altering in some sense the environment that it monitors. However, I tried to understand how big is this influnce by means of a simple test (I just don't know whether it was too simple...). Using visualvm with its default configuration (1s pooling), I noticed that the heap usage of my application oscilates in a window of aproximately 2gb. Then I changed the pooling to 10 and 30 seconds, in two following tests. With these new configurations, the heap usage is still in the same window. Am I testing this the right way? Should this change in the pooling interval change the size of the windows noticed in the monitoring? -------------------------------------------------------------- Reply to this message by going to Community [http://community.jboss.org/message/589072#589072] Start a new discussion in Performance Tuning at Community [http://community.jboss.org/choose-container!input.jspa?contentType=1&...]

Loreno Oliveira [http://community.jboss.org/people/lorenooliveira] created the discussion "Re: JBoss memory usage pattern" To view the discussion, visit: http://community.jboss.org/message/589093#589093 -------------------------------------------------------------- I tried with the JVM flag, and I'm getting the same measurements. The only difference is the time between collections. When I just watch the output of verbosegc, it takes longer to collect the garbage. When a both watch the output of verbosegc and also monitor through visualvm, the collections are much more frequent, but the usage is kept in the same window. I can say that the CPU usage is low. This is a big machine, with a lot of processors and 6gb dedicated to this vm. CPU utilization is under 10%, with some short peaks of 30%. When this snapshot was taken, I can say that 95% of the normal workload was not present, but there where at least one more monitoring tool pooling data from the application (zabbix). As it is a production server, I'm afraid of trying a memory dump. But I took two histograms of the server, one with low heap usage, and the other close to the top of the heap usage. The difference between them was char arrays, Strings, int arrays, and byte arrays. Because they are histograms, I cannot track back who is instantiating these objects. However, doing this in the same application running in the development environment and with a quite lower workload, I can see that all this stuff is being generated by the container, and not by my application... -------------------------------------------------------------- Reply to this message by going to Community [http://community.jboss.org/message/589093#589093] Start a new discussion in Performance Tuning at Community [http://community.jboss.org/choose-container!input.jspa?contentType=1&...]