Scalability Test: parallel requests not matching between app server and db.

Question

We have an environment where we have started doing some load testing on our weblogic 11g instance.
The current flow and configuration is the following:
There are https requests that go into a servlet, which then calls a local EJB which uses a connection pool to request data that is returned to the caller.
The following diagram shows the general flow w/ pool size limits:
(Socket Readers: 33%) --> Servlet --> Local EJB (initial 100, max 200) --> Connection pool (initial 400, max 400) --> Oracle RAC database
First thing worth pointing out is I know that the connection pool size is drastically larger than the ejb pool -- and this needs to be addressed.
During a scalability test, we saw that there were the following max stats on the system:
Max number of sockets to the app server was approximately 450, while the maximum number of connections used during the test was 122.
We did not have instrumentation on the ejb pool to see how much it grew, but I'm assuming that the limit was due to the pool being to small and not growing fast enough.
what are people's thoughts on what to measure and test. We are trying to get a baseline configuration before jumping to a cluster environment.
Is there recommendations in terms of industry standards for matching the socket readers, thread pool, ejb pool and connection pool for a system? General recommendation read so far indicate that inital and max should be set to the same values?
Any recommend literature is appreciated.
Thank you.

Answer 1

Another measurement point that was captured is the max number of threads (LWPs) on the server went up to 196 (then dropped to 179 and stabalized there)
Any help is definitely appreciated.