Opinion about Weblogic clustering

Hi,

I would like to hear your opinion about my clustering scenario.

I have two physical machines, both with Apache HTTP server and 2 Weblogic server istances. Apache serves static content and Weblogic dynamic stuff. All 4 servers are defined to run as one cluster. Apache server are running webserver plugin and all .do and .jsp calls are fowarded to weblogic.

The actual question is...Should I forward call from apache to all weblogic server or only to server in same machine.

Example: I have to machines A and B and they are running Apache server and weblogics A1, A2, B1 and B2. Request comes to Apache in server A, should I forward the request to weblogic A1, A2, B1 or B2 or only to A1 or A2.

Is there a performance decrease if I forward calls across different machines.

Got a few questions:
Do you have a hardware load balancer in front of the Apache Servers?
Are you using HttpSession to store user information?
Are you replicating these sessions across the cluster or are you using sticky sessions?
If you are using sticky sessions, are the sessions sticky at the load balancer or WebLogic layer?
Are these boxes co-located in the same data center?

Originally posted by Chris Mathews:
Got a few questions:
Do you have a hardware load balancer in front of the Apache Servers?

Yes

Are you using HttpSession to store user information?

Yes

Are you replicating these sessions across the cluster or are you using
sticky sessions?

Session replication is on and sticky session are used.

If you are using sticky sessions, are the sessions sticky at the load balancer or WebLogic layer?

Load balander doesn't have any idea about session so they are sticky in WL layer

Are these boxes co-located in the same data center?

Server are in same data center and in same subnetwork.

Originally posted by Tero Ahonen:
If you are using sticky sessions, are the sessions sticky at the load balancer or WebLogic layer?

Load balander doesn't have any idea about session so they are sticky in WL layer

There you go then... you answered your own question. If the hardware load balancer(s) doesn't know anything about sticky sessions then for each request the user could land on either Apache Server. Therefore, to make its way back to the proper "sticky" server Apache needs to be configured to send across the entire WebLogic Cluster.
[ April 04, 2006: Message edited by: Chris Mathews ]

That was my initial quess for the answer, but I just needed some confirmation. Thanks.