Question: Performance problem with SG with OPS edition, HPUx11i

[OSY Unix 1-4-2 (A-Team)]

Hi admins :-)

We have some performance problems during failover in ???ServiceGuard Cluster with OPS edition???...

Standard situation :
We have got a two cell ???ServiceGuard-Cluster with OPS edition??? and HPUX11.11 . On each node runs an oracle OPS instance (with shared storage/volumegroup) and also a BEA Weblogic package, both not switchable. An application sends it???s requests ??? loadbalanced ??? to the Weblogic packages, which will distribute them to both instances/nodes.

In case of failover (one node crashes and is stopped during boot process) :
Each node (with BEA/Weblogic + oracle) should be able to handle the requests without the other node. So the remaining node should do the job.

Problem :
When we switch off one node (including it???s packages) we get the following behaviour :
- every request needs ten minutes and more, normally it needed 1 second !
- one BEA connection to oracle needs one minute to establish, normally less than one second
- the failover (Application-BEA-Oracle) to the running node is very very slow

Our questions :
Has anybody ever seen or heard about or worked with such an application configuration, especially ???SG with OPS edition??? , OPS in common and BEA (Weblogic) ?
Does anybody know something about critical timeout???s or parameters in BEA or oracle ?
Have you ever experienced bad performance impact during failover for a similar architecture ?

I know : it's hard to find a similar system. But nevertheless we would be very interested in any (!!!) ideas, infos or knowhow about this topic !


With best regards, and thanks for your help !
Roland Haessner
HPUX administrator
Roland.haessner@t-systems.com
:-)