Re: Summary: Oracle and Disk Mirroring

From: Graeme Sargent <graeme_at_pyramid.com>
Date: Mon, 25 Jul 1994 14:37:45 GMT
Message-ID: <1994Jul25.143745.22915_at_pyra.co.uk>


David J. Gimpelevich (davidgi_at_rossinc.com) wrote:
: Well, I have finally got a enough responses for a meaningful summary.
 

: The consensus is, from about 3 respondents, that using the OS mirroring
: on Oracle files is just dandy and even improves performance.
 

: Two of the respondents were using Sequent and one was on Pyramid. One
: additinal repondent was using, or considering using, RAID 5 on HP9000.
 

: The Pyramid user reported some performance improvement. The RAID user
: eas anticipating degradation.
 

: Oracle personnel were remarkable by their absense...
 

: However, some questions remain:
 

: Is Oracle7's log mirroring preferrable to using OS to mirror the logs
: (independently of the rest of the files)?

I used to think so until the day we added a member to the online log group, and had to contact support to get the database back up. Maybe this was a freak event, maybe not, I've never had the time to experiment.

So now I'm not so sure.

As long as both mechanisms work reliably, it seems to me it should meke no difference, so go with the mechanism that you believe to be the most reliable.

: Can any OS mirroring system handle raw partitions?

I know Pyramid's Vdisk can. I assume others can also.

: What would happen, i.e. would the database become inconsisent in any of
: the following situations?
: sudden power failure during a write

No. Oracle would rollback any incomplete transactions on powering back up. A power failure's a power failure!

: sudden disk/controller failure during a write

In the context of Pramid's Vdisk:

If the disk failed Oracle would be unaware of the problem and would continue as normal. A message would be sent to the console informing the Sysadm that the disk had failed, so that (s)he could take remedial action at his/her leisure. Assuming the disk was in a HAAS bay, repair/replacement could be done online.

If the controller failed, assuming that the mirror was not attached to the same controller, then the situation would be as above, exept that the controller could not be replaced online. Downtime would have to be scheduled to replace the controller during a period of low activity.

If the mirror was on the same controller (Tsk, tsk. Silly!), then DBWR would abort Oracle and you would need to RECOVER the datafile(s). If the write in question was to the Redo Log (Tsk, tsk. VERY silly!) then LOGWR would abort Oracle and you would need to RECOVER the database.

: I guess I would now feel somewhat justified in recommending mirroring,
: even if somewhat guardedly.
 

: -djg
Received on Mon Jul 25 1994 - 16:37:45 CEST

Original text of this message