[jboss-user] Clustering and Loadbalancing with HTTP and EJB

[Reinhard Brandstaedter]

Hi,

I'm currently in the planning stage of a clustered JBoss Environment.
I understand that there are two service architectures for clustering and
loadbalancing. One with a client-side interceptor used for fat-clients
that communicate via ejb/rmi and a loadbalancer for HTTP access via
Apache and mod_jk.

If i need a mixed setup of both (web-access and rich-client
applications), how would someone distribute the load over multiple
cluster nodes with the following aspects:

a.) the fat-clients should have a higher priority. For example if
massive webclients cause a high load on the cluster the fat-clients
should not be influenced by this

b.) both client access (web and fat) should be distributed over all
nodes in the cluster to improve failover.

I thought of the following setup:
- 4 JBoss Nodes in the cluster partition
  (identical ressources/performance)
- 4 mod_jk connectors with different loadbalancing factors 2/2/1/1
- fat-client interceptor with loadbalancing factor 1/1/2/2

This would ensure that all nodes are utilized (for failover) and the
web-clients can't influence the fat-clients that much, right?
The main part of the web-clients would go to node 1 and 2 and only in
case of a fail utilize 3 and 4 more (maybe even use higher lbfactors
than 2).
That's the theory :-). Some Questions:

- I intend to use round-robin for loadbalancing (both clients). How can
I specify loadbalancing factors for the fat-client interceptor? I only
found a documentation how to do it with mod_jk 

- The fat-client has to download a proxy class from a cluster node. How
is this done initially? How does the client select the node? Is this a
static definition? If yes? what happens if this node is down, or does
the client need a (maintained) list of nodes (before even downloading
the proxy class) as the mod_jk loadbalancer does?

Thanks for answers!
Reinhard