Do you have another disk (controller) on which you can place the archive
redo logs?
"Marcel Duerr" <duerr.marcel_at_freesurf.ch> wrote in message
news:68830637.0209191051.5cbd20d_at_posting.google.com...
> Hi
> I need the advise of the oracle experts.
> I need to tune an application that generates about 600MB redo an hour.
> The application has a commit rate of about 25 commits per second
> (Please don't try to convince me that this is too high, commit rate is
> dictaded by application logic, I can not change it and it does not
> make sense to change it).
> Currently I have the following setup:
> Oracle 8.1.7.4.5 on Windows 2000 SP2
> Data-Files on Volume E (Striped and mirrored)
> Redo-Groups 1,3,5 on Disk F: and G: (each RAID1)
> Redo Groups 2,4,6 on Disk H: and I:(each RAID1)
> Archive-Logs on Disk J: (RAID1)
> We do force a log switch every 15 minutes (dbms_job issuing an alter
> system switch logfile).
>
> We suffer (too) high log file sync waits immediately after switching
> log-files from Disks F/G into H/I, but not after switching from H/I
> into F/G.
> Disks F/G and Disk J are connected to the same controller, so I
> believe we hit SCSI-bandwith limitations while the archiver is copying
> the last redo to J (takes about 15 secs to copy 150 mb) while LGWR
> needs to constantly write to H/I.
> Unfortunately I cannot put in a third controller.
> This leaves me two choices:
> a) Slow down the archiver to eat less bandwith.
> b) Force log switches at a higher interval, therefore reducing the
> amount of data the archiver has to copy in a chunk, but at the same
> time generating more checkpoints. As I understand, Oracle's lazy
> checkpointing makes this feasible.
>
> Finally I get to my questions:
> - can the archiver be configured to slow down? If so which parameters
> would you suggest?
> - do you think it would be ok to switch the log files every minute?
>
> Thanks for any advise.
> Marcel
Received on Fri Sep 20 2002 - 09:24:49 CDT
