You asked whether there was something you missed. Fast_start_io_target is
it. I'll happily test it for you, but my time is not free... and you
probably wouldn't want to fund the research.
HJR
--
----------------------------------------------
Resources for Oracle: http://www.hjrdba.com
===============================
"Keith Boulton" <kboulton_at_ntlworld.com> wrote in message
news:GPGe8.2517$Hg1.444047_at_news6-win.server.ntlworld.com...
> I understand what you're saying, but what I would like to see is the
results
> of actual tests, rather than what appear to be theortical statements.
>
> Howard J. Rogers <dba_at_hjrdba.com> wrote in message
> news:1014673079.588197_at_bugstomper.ihug.com.au...
> > Recovery time is a combination of the amount of redo to be read, and the
> > number of buffers to be recovered. The log_checkpoint parameters only
> > affect the first of those. You could imagine, for example, a
transaction
> > which does expensive inserts and deletes, thus generating reams of redo,
> but
> > which happened only to affect one or two buffers. Lots of redo to be
> read,
> > few buffers to be recovered. Conversely, a bunch of updates might
> generate
> > piddly amounts of redo, but dirty hundreds of buffers (each of which
then
> > has to be read from disk on subsequent startup, thus generating stacks
of
> > *datafile* I/O).
> >
> > You need to set FAST_START_IO_TARGET to limit the number of buffers
> > effectively. The_interval and _io_target parameters in tandem do a
pretty
> > good job of limiting recovery time.
> >
> > Regards
> > HJR
> > --
> > ----------------------------------------------
> > Resources for Oracle: http://www.hjrdba.com
> > ===============================
> >
> >
> > "Keith Boulton" <kboulton_at_ntlworld.com> wrote in message
> > news:IOxe8.4144$R16.741296_at_news11-gui.server.ntli.net...
> > >
> > > Keith Boulton <kboulton_at_ntlworld.com> wrote in message
> > > news:V24e8.28567$hM6.3678545_at_news6-win.server.ntlworld.com...
> > > > failure than in the last 2 hours, so I would go for a checkpoint
> timeout
> > > in
> > > > that case.
> > > >
> > > Doh!
> > >
> > > Of course, as someone else pointed out, checkpoint timeout is
irrelevant
> > to
> > > this discussion.
> > >
> > > I do, however, have a question.
> > >
> > > Has anyone actually tested the impact of these parameters on recovery
> > time?
> > >
> > > As I said, I have in the past attempted to find a penalty associated
> with
> > > larger logs, but always with checkpoint interval set high and
checkpoint
> > > timeout set to 0. While I'm happy enough to believe that my testing
> > matches
> > > by theorising about limiting factors affecting restart time, I can't
> help
> > > wondering if there's something I've missed.
> > >
> > > Any suggestions?
> > >
> > >
> >
> >
>
>
Received on Tue Feb 26 2002 - 02:56:56 CST
