Re: Database Performance Metrics - What to use?

"Jim Kennedy" <kennedy-down_with_spammers_at_attbi.com> wrote in message news:EM1ca.67048$F1.2859_at_sccrnsc04...
> One way is to have a baseline of measurements for each DB at a time when
> performance is acceptable and then compare against a base line. (eg % of
> sorts to disk is <1% for oltp, but for DW it is 30% and be concerned when
> things deviate from norm sorts to disk for oltp >3% or some such thing.)
> On the other hand, if you just want to look good you could just report
> statistics in a vacumm eg buffer cache hit ratio.

This is a particularly good one, since you can generate whatever hit ratio you require to give the precise impression that you want.

What I have been looking at is

user commits/sec - labelled "Transaction Rate" as it gives an indication of busy times
and *Database* Response time, though I'm not exactly sure how to measure this in a way I feel that I can backup.
As Tanel says Application Response time would be the better measurement, but in the absence of control of the source code for the app that becomes somewhat tricky.

--
Niall Litchfield
Oracle DBA
Audit Commission UK