undo header congestion? (8.0.5)

Hello,

I'm trying to figure out why two 8.0.5 instances are a bit slow. Both instances run on the same machine.

I had a look at stats for the rollback segments, and one of the things I've noticed is that "undo header" contention occurs quite often. Normally this happens from 0-4 times a minute, but in periods more than 1000 times a minute for one instance.

After four and a half day of uptime, v$waitstat reports this:

instance	Count	  time
instance 1	29580	  1537
instance 2	 6068	304307

(BTW, I've only monitored this for a couple of hours. Count increases steadily, but "Time" does not. I'll have more info tomorrow after some heavy batch jobs are run later today.)

The book "OCP Oracle Performance Tuning Study Guide" states on page 431 that "Ideally, the counts for these waits should be consistently at or near zero.", which definetely isn't the case for me.

Googling indicates that I shouldn't focus too much on "undo header" contention, but gut feeling tells me that something is wrong here. ;-)

Some additional info: instance 1 has two rollback segments, and instance 2 has just one (+ system).

Any comments?

Thanks in advance.
Bjørn Received on Tue Jan 20 2004 - 08:47:40 CST