Re: RAC performance and redundancy questions.

For #1

When one of the nodes goes down one of the 2 surviving nodes becomes responsible for instance recovery of the failed node. The instance that is responsible for doing the backout will be momentarily busy while it backs out any uncommitted transactions ( which might include uncommitted inserts ). Any of the database blocks that was inserted into and actually written to disk ( oracle does this at times with uncommitted changes in data file blocks ) that need backout will also get temporarily impacted.

Your DBA's should be able to help you out here and see if this is what you are talking about.

Do you get oracle error messages on inserts? Are you doing insert appends? What exactly characterizes the problem from your applications vantage point?

#2

When you are running only 1 session from your application to the database the transaction throughput will be similar whether you are selecting a specfic node or a load balanced service name. After all you only have 1 session going.

A perhaps better test might be to start 3 ( or 9 or 15 ) sessions and have them all go against the same instance then run the same sets of sessions using a load balanced service name.

Yes there is some contention between the RAC nodes but it sounds like the test you are currently running won't see that anyway if I understand you correctly. Received on Tue Nov 08 2005 - 16:57:20 CST