RE: [JBoss-dev] Clustering is available for use

[Sacha Labourey]

That is fuzzy logic of the week-end ;)   For entity beans, we let the database synchronize the beans (through select ... for update) consequently, we do not need/have a distributed cache at the application server level. Consequently, your calls can either be redirected to S1 or S2: the bean will be fetch from the database (because it needs to lock it anyway.) See the ML about previous discussion about optimist/pessimist locking at the database/AS level: some parts are interesting for clustering too. Some more features could be added in the future for read-only beans.   Remember that clustering means higher availability and in some cases better scalability but it also has a cost! Even with a distributed cache, this would have a cost.   Cheers,                                         Sacha -----Message d'origine----- De : Rick Gibbs [mailto:[EMAIL PROTECTED]] Envoyé : lundi, 22 octobre 2001 03:34 À : Sacha Labourey; Jboss-Development@Lists. Sourceforge. Net Objet : Re: [JBoss-dev] Clustering is available for use What we would like to do is the following:   From a web app request a slsb that is a facade for a local entity bean where clientid=X and ensure that all subsequent requests end up at the same jboss server.  This would allow us to take advantage entity bean caching for that client.   If we were to use RoundRobin:     request 1 for clientid=x would hit jboss1     request 2 for clientid=x would hit jboss2 now we would have 2 copies of the same entity bean on 2 seperate servers unless I am missing some core concepts.   Did that make sense or is my logic fuzzy this weekend.   Rick ----- Original Message ----- From: Sacha Labourey To: Rick Gibbs ; Jboss-Development@Lists. Sourceforge. Net Sent: Sunday, October 21, 2001 6:51 AM Subject: RE: [JBoss-dev] Clustering is available for use Hello Rick,    Which kind of bean are you using i.e. what is the target of your clients? If they are SLSB, round robin is used. For SFSB (not yet available, under development), calls for a same SFSB will always be targeted at the same node and, if this node dies, be redirected to a (number of backups is user-definable) backup node that will have its last known state.    In the future, we may imagine to have load-balancing policies user-definable through jboss.xml (not hard to do). In the meantime, take a look at org.jboss.ejb.plugins.jrmp13.server.JRMPContainerInvokerHA (in the cluster module)   Cheers,                                         Sacha     P.S. Home calls (create) are always load-balanced I was wondering what the easiest way to define custom load balancing polices would be.  We have an application which hosts multiple clients and we would like to develop a policy that ensures all requests for a particular client goes to the same server for each request so the their data is cached in one place.  I think it would be farily easy to implement on top of a RoundRobin policy that exists just not really sure where to start.  Any pointers?