I am quite a jboss newbie, dealing with a commercial website made of :
1 f5 load balancer
2 apache with mod_jk 1.2.26 ( with load balancer set as busyness )
4 application servers ( all of them lbfactor of 1 ) 12gb ram, 8cpu, java 22.214.171.124 and jboss-eap-4.3.0.GA_CP06 ( -Xms6144m -Xmx6144m -XX:MaxPermSize=1024m )
2 oracle boxes set up with RAC by dba on the backend
They are running fine and the website is looking good. Course sometimes they swap but nothing to worry about.
I have been told that in a couple of weeks we are gonna have an increment in the accesses on the website.
I do have cacti to monitor apaches and jboss servers ( httpd threads, processes, memory, cpu, load, network et similia ).
Apache is just fine, I cannot see any bottleneck in it at the time being. What I am concerned about is the app servers. By having a look at cpu, load, memory, network they look fine, but what I need to find out in time is whether or not they will be coping with the growing traffic.
Is there anything particular I should be looking at ? I need to make some assumptions before we get the increase of traffic, just to be safe.
What data are you tracking that will tell you the load on the system? What kind of load testing have you done? Without any idea of how "busy" the app server systems are, it is hard to say.
Also, the systems have too much RAM if the heap is maxed out at 6GB + 1GB permgen - 8 GB RAM would suffice. But hey, you have room to increase the heap. And if you are using the CMS collector the more heap the better. (What collector are you using? Full list of JVM options would help.)
Are you applying any affinity on the app server boxes or are you letting the JVM run wild over all 8 cores? Have you done any load testing to determine if 2, 4, 6 or 8 cores is the best number to handle you app? In some testing I have done we actually gained performance by reducing the number of cores from 4 to 2 in some situations (we did this via infinity settings).
By the way, I doubt that we can help that much - this is usually a problem for high-paid consultants who will come in and monitor and tune the system and provide recommendations on how to handle increased load.