Re: Question about sizing log files

The proper sizing of redolog files is a trade-off between better runtime performance and better recovery performance.

Bigger log files --> less frequent log switches --> better runtime performance.

Bigger log switches --> less frequent checkpoints(provided you have high log_checkpoint_interval) --> more time required during instance recovery.

Smaller log files --> frequent log switches --> poorer runtime performance.

Smaller log files --> frequent checkpoints --> less time required for instance recovery.

• a checkpoint occurs at every logswitch.

HTH,
Sanjay

Chuck <ccarson_at_echeeba.com> wrote in message news:<3E8B6FDA.8020607_at_echeeba.com>...
> Hans Forbrich wrote:
> > Chuck wrote:
> >
> >
> >>We currently have our redo logs set to 20mb and we get an average of 4
> >>log switches per minute. Just wondering how high people tend to set redo
> >>log sizes on very busy OLTP database. We are currently running 8.1.7.4
> >>on Solaris 8 64-bit. We are currently trying to migrate to 9.2.0.1.
> >>
> >
> >
> > This becomes a 'philosophy' question ... basically how much paper backup do you
> > want to keep in case the systems fails. It's really a balance between cost of
> > data protection and cost of CPU cycles to accompdate the load for high data
> > protection.
> >
> > Generally I try to configure the redo logs to switch for 10-15 minutes at 'normal
> > busy load'. In some cases this has required log files of 150 or 200 MB. You
> > indicate 80Mb/minute or roughly 800Mb in the 10 minutes I'd target; that's
> > slightly high for my taste but not totally unreasonable. Flip side, depending on
> > the speed of disks, archiving the 800Mb logs might cause a noticable stutter in
> > which case I'd probably head back down to 150-200 again.
> >
> > Are you getting that 80Mb/min rate continuously? Or is that a peak load? (If
> > continuous & it's an order entry system, I want the stock-ticker!)
> >
> >
> >
>
> *lol* Yea I too wish it was an order entry system. Actually I am at a
> Bio-tech company and the data is bio-tech related. We have a lot of
> automated processes, primarily a 500-node linux cluster that hammers out
> a ton of data.
>
> My main reason for increasing the interval betweem log switched is to
> make 'life easier' and reduce the quantitity of archive logs.
>
> -Chuck
>
>
>
> -----------== Posted via Newsfeed.Com - Uncensored Usenet News ==----------
> http://www.newsfeed.com The #1 Newsgroup Service in the World!
> -----= Over 100,000 Newsgroups - Unlimited Fast Downloads - 19 Servers =-----
Received on Wed Apr 02 2003 - 22:00:55 CST