Architecture for using MDB with a server cluster

Question

          Hi,
          I would like to get some advice on the architecture that is the most desirable
          for a scenario that I have here.
          I have 3 machines, each running a managed server belonging to the same cluster.
          My admin server runs on a 4th machine. I have deployed my MDB on the cluster and
          I'm trying to find a way to configure the 3 + 1 machines to give the best end-to-end
          time for processing the JMS messages.
          To investigate this, I'm running some simple benchmark tests with a client application
          that sends JMS messages to the system. The client repeatedly sends messages to
          the MDB's queue and the MDB puts processed messages to another queue that the
          client is listening to.
          Now, my question is: where should I set the target for my JMS server (and hence
          my queues) and the connection factories? I can think of a few possiblities:
          1. Connection factories target: cluster; JMS server target: one of the servers
          in the cluster
          -> Potential drawback - The server with the JMS server will be handling and redistributing
          the JMS messages to other servers in the cluster. This means that a portion of
          its processing power is used to do this instead of actually having the MDB process
          the JMS message. (Please correct me if i'm wrong)
          2. Connection factories target: cluster; JMS server target: have one JMS server
          for each server in the cluster and make use of distributed destinations
          -> Potential drawback - My client establishes connection with the MDB's queue
          only once before it sends its messages to it. Probably as a result of this and
          the way WebLogic clusters load-balance themselves, all the messages end up being
          routed to the same server. This option appears to be out since 2 of the 3 servers
          are not utilized at all.
          3. Connection factories target: admin server; JMS server target: admin server
          -> Potential drawback - The MDB has to maintain a queue connection with a server
          that is not part of the cluster. (Again, please correct me if I'm wrong.) I'm
          not sure if this introduces extra time taken for the MDB to receive its messages
          and for it to send the processed messages to the queue.
          I'd appreciate it if someone could advice me on the most desirable architecture
          to use here. From my understanding of the problem, option 3 seems to be the answer,
          but I may be wrong. Perhaps there is no significant difference in terms of performance
          that 3 can give, compared to 1 and 2.
          One last question. I'd like to understand, in option 1, if the admin server plays
          any part in load-balancing the cluster. Are the JMS messages received on the cluster's
          JMS queue forwarded to the admin server before they are rerouted to the server
          that is supposed to process it?
          Cheers,
          C.Y.
          

Answer 1

3. Connection factories target: admin server; JMS server target: admin          server
          > -> Potential drawback - The MDB has to maintain a queue connection with a
          server
          > that is not part of the cluster. (Again, please correct me if I'm wrong.)
          I'm
          > not sure if this introduces extra time taken for the MDB to receive its
          messages
          > and for it to send the processed messages to the queue.
          Admin server is not supposed to participate in the cluster. I wouldn't
          deploy anything on the admin server.
          I think my personal preferene would be connection factories to the cluster
          and use distributed destinations.
          Regards...