Re: Slow OLTP users 14CPU SunFire

Your box appears to be way over sized. For 100 commits per sec, you certainily do not need 14 CPUs.

I would seriously look into lgwr. What is your average redo write time? What kind of disk is it on?

Your redo must be severely out of tune to cause sub second transaction to be over 5 seconds. I would switch to raw device for redo that would cut down 50% of response time and lgwr CPU right away. Also if you are using buffered IO (file system) plus bunch of members the response time would be worse. I would target average redo write time to be < 3 MS if you are on SAN - if it is tunred right.

Also just to make sure you are not over using degree of parallelism for the table creation, you should do a vmstat to check you CPU utilization.
Because if you do abuse parallel query, one single SQL can kill the entire
box even though you have 14 CPUs. You should always know where you are standing on CPU % busy and IO service time, before digging into Oracle for performance problem Received on Fri Jun 03 2005 - 12:37:22 CDT