Re: Redo Log Size

The size of logs is determined by the rate at which you wish to log switch (and hence, to checkpoint). Small logs fill and switch faster than big logs. Personally, I recommend humungous logs that never switch (until you force a switch at 2am when no one gives a damn). But many DBAs seem to think that a log switch every hour or so is acceptable. (The trade off is Instance Recovery time -Instance Recovery means re-perform all transactions since the last checkpoint... if you're not checkpointing very often, then that could be a lot of transactions to replay).

Media Recoveries are not affected by the size of logs, since Media Recovery means replay all transactions since the last backup -and it doesn't matter if those transactions fill 3 archives or 30.

The number of groups is determined by the 'loop back' issue: You don't want to switch back to the first log group until it has been successfully checkpointed and archived, because there'll be a hang if you try it (you can't over-write active redo logs, and they stay active until checkpointed and archived). To avoid switching back on top of yourself, therefore, have extra groups.

The number of groups has absolutely zero bearing on recovery.

Regards
HJR "Susanne Bagl" <susanne.bagl_at_chello.at> wrote in message news:YKTi7.36177$Xd1.600693_at_news.chello.at...
> Although I fully agree with you there is still an interesting question for
> me:
>
> Is it better to have larger log members instead of more log groups? Or
isn't
> it relevant if there is enough time for archiving before the log-switch
> cycles? Does it matter in case of recovery?
>
> Susanne
>
> "Suzi" <nospam_at_nospam.com> schrieb im Newsbeitrag
> news:fJRi7.22832$6x5.5027620_at_afrodite.telenet-ops.be...
> > 1M .... large ... ??? you mean small !
> > 4 logswitches in less than one minute - why don't you make'm 30MB or
> > something ... should give
> > you a logswitch every 10 minutes or so ... still pretty high but better
> ...
> > We have a 800GB DB with 12 loggroups and each member is 256MB so you
> > shouldn't loose your
> > sleep if you think 2M is too large ... and oh ... as always tune your
> > checkpoints.
> >
> > Who installed your DB - if I were you I would check if I didn't have a
> 100k
> > logbuffer and an 8Mb buffercache ... smells
> > like a default install ...
> >
> > "Bryce Moulton" <brycemoulton_at_eaton.com> wrote in message
> > news:9mgl8q$eek2_at_interserv.etn.com...
> > > I have a 16G db that houses BLOB's anywhere from 0 to 2M.
> > > I have 4 redo log groups with each member being 1M in size.
> > > 1M seems to be pretty large, but I'm getting errors in my alert log
> > because
> > > arch isn't finished archiving when the log group needs to be used
again.
> > > Sometimes in the alert log I can see where all 4 of the log groups
have
> > been
> > > gone through in less than 1 minute.
> > >
> > > I believe all indications point to increasing size or # of groups, but
> I'm
> > > not sure if Oracle recommends a max size for redo log files.
> > >
> > > Should I increase the size to 2M, or perhaps the # of groups to 5 or 6
> > > groups??
> > >
> > > I would greatly appreciate help. I can't seem to find much
documentation
> > on
> > > this.
> > >
> > > Thank you, Bryce Moulton
> > >
> > >
> >
> >
>
>
Received on Wed Aug 29 2001 - 03:53:16 CDT