Re: Improper Load Balancing

Donald

It may not be relevant in your environment, but we had similar problems in a 
three tier system (OC4J webserver, OC4J app server, RAC database, 9iR2). The 
problem was that all app server nodes had the same configuration - so they all 
tended to connect their first N sessions (the session pool minimum) to the same 
node. You could be seeing the same thing; web clients are nicely load balanced, 
but the web-to-DB connection isn't. When the session pool is created, there's 
no load on the RDBMS - so a connection storm (like server startup) tends to hit 
one node - load metrics are only updated slowly (after a few seconds at least) 
- so SQL*Net load balancing just isn't suitable (at least at 9iR2 RAC) for 
session pools; you have to help it along.

More background on what we did to work around this and a related problem (if 
you are interested) here: 
http://preferisco.blogspot.com/2006/04/micro-partitioning-pooled-sessions-in.html

Regards Nigel


----- Original Message ----
From: "Freeman, Donald" <dofreeman@xxxxxxxxxxx>
To: oracle-l@xxxxxxxxxxxxx
Sent: Thursday, March 22, 2007 12:55:16 PM
Subject: Improper Load Balancing



We are experiencing problems with the database connectivity from the production 
web servers for the last 2-3 weeks. We are using Oracle RAC in the production 
and there are 2 database servers. Somehow the connection goes to one database 
server intermittently, not all the time. But when it happens we see most of the 
connections going to the same Oracle node i.e. second node. We are using 
Windows load balancing on the web servers. Does anybody have any idea what 
might cause this?
Don Freeman 
Database Administrator 1 
Bureau of Information Technology 
Pennsylvania Department of Health 
(717) 703-5782

Other related posts: