This content has been marked as final. Show 4 replies
Depends on how much memory a node has.
"...currently we care running managed server with 8gb jvm heap..."
You have to take into account the work memory of JVM as well.
"...What is the best practice for getting optimal performance."
When you have an 8GB heap with a lot of live objects, a full collection can degrade your performance drastically.
Of course exalogic is a big machine with lots of CPUs, so you have to choose the right garbage collection algorithm
that can take advance of this (concurrent and parallel collectors).
Some more info on JVMs can be found here: http://middlewaremagic.com/weblogic/?p=6930 and http://middlewaremagic.com/weblogic/?p=7083
technical session - http://middlewaremagic.com/weblogic/?p=7831
Thanks for the reply René..
We have 96GB RAM and 24 processor on each node..
Is it good to have some 8 managed server's per machine which can result in around 240 (30*8) JVM's in the cluster... is it good idea to have such a huge cluster of managed servers? Our current (non exalogic) app uses only 8 managed server's.
Also if you can suggest if 8 managed servers can scale on above specification?
If you want to go for cluster of 240 servers you need to use multicasting instead of the default unicast.
But what would be the reason for 240 servers?
You can always start-out with a cluster of 8 servers and add servers when needed (this is a feature that WebLogic has in it self)
- you can add different machines on which you add servers
- or you can add the servers to existing machines
An example is given here: http://middlewaremagic.com/weblogic/?p=7795 (look in the loadbalancing section)
Note that when you add servers to existing machines you have to keep in mind to balance the number of CPUs and JVMs
about 2CPUs per JVM will usually do fine.