From: Theresa Sarver (IFMC.tsarver@SDPS.ORG)
Date: Mon Oct 07 2002 - 14:39:28 EDT
I found this post (and it's resolution) most intesting as Holger's problem is almost identical to a problem that I am having on one of our SP [Wide] nodes.
SP 9076 - this particular node has 4-375Mhz processors w/ 4GB RAM
AIX 4.3.3 ML8
Oracle 8.17
About 3 weeks ago the DBA's migrated 'a lot' of oracle data onto one of our production nodes and since then things have been out of control! On average the run queue now has 5-10 jobs in it at all times (though I've seen it up over 37 at one point!) The buffer has 2-4 jobs in it. There's very little paging going on. usr has most of the CPU tied up at approx 65%, and sys is about 25%. The server is about 20% idle with anywhere from 2-11 jobs in the wait queue.
Oracle's load is spread across approximately 65 processes, these processes though it is rare to see them above 8-9%, do go over 22% on occastion. Also, the tracedata that I've been running (tprof/filemon) shows a classic I/O bottleneck with some of the Oracle files that were recently uploaded transferring up to 220MB/s. One other thing - I noticed from my sar output that there are a lot of page faults - thus causing an incredible amount of disk I/O.
The first corrective measure I tried as creating 4 new mount points and then having the DBA's move those busy tablespaces off the hardest hit logical volumes, however, that has failed to show any signs of improvement.
It almost appears that there just isn't enough processor power in the node to be able to handle this increased CPU load. ??? And thanks for Holger's post I now have one other avenue to explore...so I'll also go back and talk to the DBA's again and see if there might be an indexing problem.
If anyone has any other insight into what might be going on - I'd welcome the help...otherwise, thanks again for the post/followup Holger!
Theresa
>>> Holger.VanKoll@SWISSCOM.COM 10/07/02 10:13AM >>>
solved, a db-admin deleted an index. recreated it, ok.
I dont understand why this lead to this system-behaviour; but oracle is
another world...
This archive was generated by hypermail 2.1.7 : Wed Apr 09 2008 - 22:16:15 EDT