5 Replies Latest reply on Mar 1, 2007 11:14 PM by schustes

    loadbalancing without ha-failover

    fwiese

      Hi,
      is it possible to use the jboss loadbalancing without the high availability failover features. We can't use the default stateful session bean replication mechanism between cluster members. It's not so important to have high availability. Our preference is system performance. If our customers have perfomance problems, we want to install additional server instances and so on.

      In the moment we have a mixed stateful/stateless architecture, and we are at the border of operating system management (2 GByte RAM for each process in 32-Bit architectures). Exists their a possiblity to realize this requirement?

      Thanks for your help, Falko Wiese.

        • 1. Re: loadbalancing without ha-failover

          Just don't set the clustered tag true in your jboss.xml for your SFSB. It will just do the load balance.

          • 2. Re: loadbalancing without ha-failover
            fwiese

            Thanks, this is a solution, but -- exists a global switch also ;-), it would be nice. A thing like a default setting for all SFSBs.

            Falko

            • 3. Re: loadbalancing without ha-failover
              slaboure

              you could also try to make your own clustering configuration, name it the way you want, and simply use a non-clustered cache/store, like for the default SFSB configuration.

              Cheers,


              sacha

              • 4. Re: loadbalancing without ha-failover
                schustes

                What am I doing wrong then, or does my scenario just not fit. The application I have is basically a distributed computing server used to run simulations requested by a remote client. I also want to have the loadbalancing, but no replication, so I use HA-JNDI, but set don't set the clustered tag in jboss.xml

                The architecture is quite simple - each node has the SFSBs to be balanced. A central node is dedicated to serve remote client requests and has the single-cluster jms running. There are two ejbs on the normal nodes, one being a proxy, creating n delegates to split up the workload. Thus, the central server creates an SFSB on a randomly selected server, from where possibly more SFSBs, which should be loadbalanced, may be created.

                My problem occurs when I want to create 'delegate beans' on the nodes. I start a lookup, and what (probably correctly) happens, is that the Home interface seems to be returned from the local jndi-tree, and consequently the beans as well. Thus, a single job will run only on one node (the one selected by the 'bootstrapping server') with the proxy and a number of delegates it creates. What I would like to have is to distributed the delegates over the cluster. Does this scenario simply not work if all nodes have the same EJBs deployed? Must I then use the clustered-invokers and add my own interceptors (http://www.jboss.com/index.html?module=bb&op=viewtopic&t=80210 - this works fine if I declare the ejbs clustered, but is just unnecessary overhead for my purposes).


                "sacha.labourey@jboss.com" wrote:
                you could also try to make your own clustering configuration, name it the way you want, and simply use a non-clustered cache/store, like for the default SFSB configuration.


                I don't know what this means. I played around with the clustered-ejb configuration in standardjboss.xml trying to disable the state-transfer, but I didn't get it to work (once I stop using StatefulHASessionSynchronisationInterceptor, the session seems to get lost, and I get remote exceptions). If there is a wiki or something, I would be very thankful for a pointer.

                best, stephan

                • 5. Re: loadbalancing without ha-failover
                  schustes

                  P.S. I use ejb 2.0 with jboss 4.0.2