Re: Lose statistics on instance bounce?

Are you looking at hit ratios since startup during that initial 4-5 days? If so, try taking a snapshot over a smaller amout of time using either utlEBstat or your own CACHE tool. That is, after the first several hours  or 1 day ) after a bounce of the instance, take an hour long sample during a busy time and calculate your hit ratios off of the sample.

Bouncing the instance certainly clears the buffer cache AND all of the v$sysstat "statistics", but it does NOT clear table/index statistics generated by calls to ANALYZE.

-Kevin

"Speed Razer" <speedrazer_at_NoT_SpamM_usa.net> wrote in message news:4j1uvs8o380b9h8pr2s7jioc9vsrvkp7ue_at_4ax.com...
> Some time back I remember reading a post on Metalink by some guy
> claiming that he preferred hot backups to cold backups because when
> you bounce your instance for cold backups, you lost all your
> statistics. I ran that concept by some other DBAs who said that was
> hogwash. However, I've seen what I believe to be exactly this
> behavior. Namely, if I bounce the instance and don't analyze
> statistics, it takes 4-5 days for the db buffer cache hit ratios to
> climb above 90%. But if I analyze right after I restart the instance,
> the hit ratio is usually up to 99% within a day. Any thoughts??
>
> Many thanks,
> SR
Received on Tue Oct 31 2000 - 12:40:51 CST