Discussions

Performance and scalability: Java Client accessing WebSphere Horizontal Cluster

  1. Hi,

    I have created a Horizontal WebSphere Cluster with Thick Servlet Redirector. Life is good as long as I access EJBs through servlets.

    When I access EJB through Java Client I face problem.

    This works good if both WAS are running and I can use any machine name as the Provider.

    Now, when I stop 2nd WAS machine while 1st running it still works, but if I stop 1st WAS machine while 2nd running it won't work.

    (NOTE: I started clustering process on 1st machine so it might have some internal dependency ???)

    Any help/clue/suggestion greatly appreciated.

    Thanks in advance...


  2. You need to process your client jars using wlmjar. If all your beans are in the same package in the same jar then this works well. If not then manually create the stubs using a combination of rmic and wlmsg.

    Once you have the stubs then the next step is to create a model and clone it on the server.

    Create a model by right clicking on the server and click create model. Check both check boxes (recurse and make a clone). Click ok. Now you should see a model in the tree view and create another clone by right clicking it and select create clone. Pick the node you want the clone on and then ok.

    You need to manually copy the files (deployed jars, servlets, jsps, images) etc to each node you make clones on. For development, a shared filesystem works, for production use replication with maybe IBM AFS or a similar solution.

    After that you should be fine. Whilst it's possible to create a model from a container this DOES NOT work. Only create models from servers.