Re: [oracle-L] [Long] How deal with load-average-spike ?
- From: "Dennis Williams" <oracledba.williams@xxxxxxxxx>
- To: goodjobfastcar@xxxxxxxxx
- Date: Sat, 16 Dec 2006 13:55:47 -0600
Peter,
It sounds as if you have an excellent start, very thorough. Two things that
have worked for me:
1. Don't neglect what is happening at the server level (and you haven't).
2. It is easy to script the ps command so you get a running record of
activity over time. A cheap tool is to format the ps results with perl, then
toss the results into MS Excel so you have a graph to show management
(managers are attracted to graphs), but seriously stuff will jump out at you
in a graph.
3. STATSPACK snapshots are excellent (and free) for gathering a permanant
record of what was happening during the spike within the instance.
You may also want to include a section about "now that I know what is
happening, what can I do about it?"
As you describe, a sheer spike in transaction volume may overwhelm the
database. A classic in the web world is for marketing to issue a special
coupon and not think to notify I.T.
But the Oracle database is a robust engine that can often cope with
incredible volumes of transactions. Often, careful analysis reveals another
database job or report or server process that disturbs the instance
equilibrium. Maybe forces the "hot" blocks out of the buffer cache.
Sometimes the answer is to reschedule that "other" job. I've even had the
experience once the job is found, for someone to say "hey I didn't think
that thing was still running".
Good luck, please post a copy of your completed paper.
Dennis Williams
Other related posts:
- » Re: [oracle-L] [Long] How deal with load-average-spike ?
