1 Reply Latest reply on Dec 24, 2004 8:42 AM by starksm64

    HTTP session replication delay

    rahulkausale

      Hello all Jboss gurus out there,

      I am having a development environment with following details,

      I have two intel p4 machines with 2 GB RAM, RedHat Linux running on it and having 3 Jboss-4.0.0 instances.

      Machine 1:

      Instance 1: running in "all" mode and having "DEVPartition" as cluster partition. in this partition only HTTP session replication is enabled.
      Instance 2: running in "standard" mode, running on different ports using binding manager service.
      Instance 3: running in "standard" mode, running on different ports using binding manager.

      Machine 2:

      Instance 1: running in "all" mode and having "DEVPartition" as cluster partition. in this partition only HTTP session replication is enabled.
      Instance 2: running in "standard" mode, running on different ports using binding manager service.
      Instance 3: running in "standard" mode, running on different ports using binding manager.

      Only "Intance 1 " on both the machines are forming cluster, and replicating session. And I am using apache2+mod_jk2 for doing the load balencing. mod_jk2 is configured to persist to same server for "instance2" and "instance3" (session sticky), but would do load balancing for the requests to "instance1"

      The problem I am facing with is, the session replication takes some time (i don't have the exact statistics) to replicate the session across the nodes, which is causing the application to fail while looking for the session it has created on one node and the request is being sent to another server by mod_jk2.

      I think this latency is byproduct of jboss serializing the session data before replicating on another server and the network latency. I don't think the network latency should cause much botheration, as both the nodes on same physical 100Mbps LAN.

      If anybody has faced the same problems or got solution for this, please help me.

      Thanks a lot

      Rahul D. Kausale

        • 1. Re: HTTP session replication delay
          starksm64

          The default replication mode is REPL_SYNC as configured in the deploy/tc5-cluster-service.xml, so unless you have multiple requests for the same session bouncing across nodes, there should be no delay issue.

          Note that the current session replication implementation cannot be used for load balanced requests. It can only be used for failover.