Re: redo logs tuning?

Well, checkpoint always occurs during a log switch, so it all depends on the size of your log files, then. Recovery won't take that long if you keep the size reasonable, so that log switches will occur once or twice per hour under normal load conditions.

The 512 bytes I "usually" see has to do with the lack of experience as DBA; never worked anything else than NT or AIX (and very little OpenVMS). Sorry for that - it *is* OS dependant, indeed.

--
Frank
Howard J. Rogers <howardjr_at_iprimus.com> schreef in berichtnieuws
39b63157_at_news.iprimus.com.au...


>

> "Frank" <frankbo_at_interaccess.nl> wrote in message

> news:8p2gnj$kn4$1_at_porthos.nl.uu.net...

> > I beg to have a different opinion:

> > set log_checkpoint_interval so high, that no checkpointing will occur,

 other


> > than during log switches. This is also the reason you want the timeout

 set


> > very high,

> > or disable it completely.

> > Check the tuning manual on tuning check points (Ch 15), gents.

> >

> > As for the size, check how often switching occurs during "normal"

 operation


> > of the database (see your alert log file). Aim for 1 or 2 per hour.

> >

> > log_checkpoint_interval is in OS blocks (usually 512byte), indeed.

> >

> > Open to other opinions - always enjoy a good discussion!

> >

> > --

> > Frank

>

> I agree Frank: the best performance option is *never* to checkpoint during

> the day, and to have one mammoth I/O fest at the dead of night when no-one

> gives a damn.  However, the corollary of that approach is enormous

 Instance


> recovery times, and that might not be acceptable to many people.  Mean

 Time


> to Recover is not something that can lightly be ignored on most systems.

>

> As an aside, the "usually 512byte" is a bit of a whitewash: it really is

> platform specific, as Steve advises, and his web site (which no

> self-respecting Performance Tuner can afford to ignore) is entirely

 correct.


>

> Regards

> HJR

>

>

>

>

>

> > Steve Adams <steve.adams_at_ixora.com.au> schreef in berichtnieuws

> > 39b4b9fb.104161686_at_nsw.nnrp.telstra.net...

> > > Hi All,

> > >

> > > No, the units for the log_checkpoint_interval are log blocks. The log

 block size


> > > is platform specific. It is often 512 bytes, but by no means always.

 The
 "Why


> > > raw log files?" tip on the Ixora web site contains a table of the log

 block


> > > sizes for the most popular platforms.

> > >

> > > @   Regards,

> > > @   Steve Adams

> > > @   http://www.ixora.com.au/

> > > @   http://www.christianity.net.au/

> > > @

> > > @   Going to OpenWorld?

> > > @   Catch the Ixora performance tuning seminar too!

> > > @   See http://www.ixora.com.au/seminars/ for details.

> > >

> > >

> > > On Mon, 4 Sep 2000 19:05:01 +0200, "Sybrand Bakker"

 <postbus_at_sybrandb.demon.nl>


> > > wrote:

> > >

> > > >

> > > >"Niall Litchfield" <n-litchfield_at_audit-commission.gov.uk> wrote in

 message


> > > >news:8ovunp$2b7$1_at_soap.pipex.net...

> > > >> "Sybrand Bakker" <postbus_at_sybrandb.demon.nl> wrote in message

> > > >> news:968007698.22997.0.pluto.d4ee154e_at_news.demon.nl...

> > > >>

> > > >> > Also, don't forget to check log_checkpoint_interval in

 init<sid>.ora.
 In
 the


> > > >> > 5M situation it should be 10000.

> > > >>

> > > >> Presumably for a 2k Block Size? If not how did you arrive at this

 figure?


> > > >> --

> > > >> Niall Litchfield

> > > >> Oracle DBA

> > > >> Audit Commission UK

> > > >>

> > > >>

> > > >AFAIK this is in 512-byte units.

> > > >

> > > >Regards,

> > > >

> > > >Sybrand Bakker, Oracle DBA

> > > >

> > > >

> > > >

> > >

> >

> >

>

>