Re: Q: Filesystem choice for log_archive_dest

I don't think buffered I/O is going to help you. For any files associated with the database, including archived redo logs, you want the data written to the disk (physically or in a protected cache) when the OS tells Oracle that the I/O write request is complete. Even with a buffered file system, Oracle will buffer the reads, but "write through" the buffer for writes to make sure the data is really on disk.
 Find out what type of disks the archivelog dest file system is built on. Then work with your SA or storage engineer to provide you with dedicated disks, add spindles, or move to a RAID implemention that handles writes more efficiently (say RAID 5 -> RAID 1+0).
 John Smiley
Technical Management Consultant
TUSC, Inc.
 On 6/23/05, Radoulov, Dimitre <cichomitiko_at_gmail.com> wrote:
>
> Thanks,
> the problem is that now I'm working on a system with direct IO activated
> on the log_archive_dest filesystem (the SA told me that this was the
> workaroud for previous "performance problems")
> OS: Solaris 8
> Oracle: 8.1.7.4 <http://8.1.7.4/>
> OFA(including log_arch_dest): vxfs fstype
> mount output:
> /smsp_ofa on /dev/vx/dsk/osmspdg/smsp_ofa
> read/write/setuid/mincache=direct/convosync=direct/delaylog/largefiles/ioerror=mwdisable/dev=449022b
>
> ... and it's now that the arch processes are really not able to keep up
> with the log file switches: "cannot allocate new log" messages + shutdown
> abort from the SUN Cluster fault monitor for timeout ...
> So I was wondering, could the direct IO be beneficial for the
> log_archive_dest filesystem in some cases?
> Cheers
> Dimitre
>
> ----- Original Message -----
> *From:* John Smiley <jrsmiley_at_gmail.com>
> *To:* cichomitiko_at_gmail.com
> *Cc:* oracle-l_at_freelists.org
> *Sent:* Thursday, June 23, 2005 5:36 PM
> *Subject:* Re: Q: Filesystem choice for log_archive_dest
>
> As long as your archiver processes are able to keep up with the log file
> switches, it doesn't matter what type of filesystem they're on. If you start
> seeing the archiver lag behind, or worse, you system comes to a screeching
> halt because it can't perform a log switch, you can increase the number of
> archive log processes (if the disks supporting the log archive dest(s) can
> handle more load) or store the archive logs on faster I/O. Typically this
> takes the form of LUNs on a RAID 1+0 using your file system of choice. For
> best performance, give the archivelogs their own set of disks.
> John Smiley
> Technical Management Consultant
> TUSC, Inc.
>
> On 6/23/05, Radoulov, Dimitre <cichomitiko_at_gmail.com> wrote:
> >
> > Hi,
> > could anyone provide guidelines for the log archive destination
> > filesystem's/IO type choice (both MS and *nix solutions would be
> > interesting)?
> > Thanks,
> > Dimitre
> >
>

>

--
http://www.freelists.org/webpage/oracle-l