we are having a problem in our clustered environment during performance testing.
WE are testing 150 concurrent users in the following environment:
1 load balancer
2 web servers
2 app server instances (each on its own solaris box)
1 admin server (located on the APP server 1 box)
Each solaris box has 2 900mhz CPUs and 4GB of memory.

during the testing, the following occurs:
1. system CPU usage on app server 1 box goes way up and redlines the system
2. user cpu usage goes down on app server 1 box
2. system and user CPU usage goes way down on App server 2 box.

we know the memory usage on the app server instance is ok, from looking at the

however, what we are concerned about is that the admin server instance is running
out of memory. we have it set to -ms64m -mx64m. Unfortunately, the admin server
logs were not captured, and we cannot repeat the tests for a while. so, in the
mean time, I pose this question:
1. if the admin server ran out of memory, would you expect to see behavior similar
to what I described above?
2. what is the current thinking on how much memory should be allotted to the
admin server?