Re: Oracle Stats

From: joel garry <>
Date: Mon, 18 May 2009 11:03:04 -0700 (PDT)
Message-ID: <>

On May 17, 2:08 pm, wrote:
> On May 15, 2:03 pm, ddf <> wrote:
> > What you posted doesn't report the time between redo log switches.
> > Try this query and see what it reports:
> > SELECT ROUND(AVG(1440 * (b.first_time - a.first_time)), 0) "Log switch
> > time - minutes"
> > FROM v$loghist a, v$log b
> > WHERE b.sequence# = a.sequence# + 1
> > AND a.sequence# = (SELECT MAX(sequence#) FROM v$loghist)
> > ORDER BY a.sequence#;
> > I'll bet it returns a value greater than 0.
> > David Fitzjarrell
> Query returns this:
> Log switch
> time - minutes
> -------------------------
>                        25

That's good. I had in mind a slightly different question, though, and that is not what is the average, but do you have times when switches are far greater than this reasonable average, particularly when you are running the tool in question. If you don't, then that is also interesting, including the point that having all those waits with no complaints may mean nothing is wrong, or may mean you are setting up for a slight I/O bottleneck cascading into a serious one after a slight change in app performance. If the latter, there may be things you can do to move more of the load onto those idle cpu's.

Please research how to see what your log buffer size is. I won't go into it now, but I will caution you it is not likely to be what your init file says, if anything. The term "granule" is important. Also, give us a clue what kind of disk configuration you have - where your redo logs are, where your data files are, do you have everything on one disk/controller...

Some clue as to what the app is doing might help too.


-- is bogus.
Received on Mon May 18 2009 - 13:03:04 CDT

Original text of this message