RE: Separating online redo logs from database files
- From: "Mark W. Farnham" <mwf@xxxxxxxx>
- To: <vbarac@xxxxxxxxxxxx>, <oracle-l@xxxxxxxxxxxxx>
- Date: Fri, 9 Apr 2010 14:39:29 -0400
Does your configuration support cache fencing? If so, and if there is a
concern that actually writing the redo logs (LGWR) is a bottleneck to
transactions, then separate volumes *may* allow you to make sure there is
sufficient write cache available to buffer your redo write peaks, even if
there is an underflow in cache availability overall to your SAN, and even if
your overall storage is S.A.M.E. Actual load conditions would dictate
whether that is an economic use of cache, presuming your hardware and
software even allows you to do this (and figuring in any software upcharge,
if any, to do this on your particular SAN).
You mentioned EMC directly. Is that RAID-1 or RAID-1 on top of RAID-S or
what?
Tim Gorman's entry in this thread was also interesting. He did a
particularly good job of listing the reasons why it might not be worth the
bother.
Finally, if your SAN is overdriven and your redo log writing is a big part
of the load, there may be price points at which it is cost effective to
de-heat your SAN by moving redo to something else, like SSD, if it can be
configured such that its i/o pathway is non-interfering with your SAN. But
you don't really need to do that preemptively, because it is so easy to
relocate the online logs simply by defining new ones and deleting the old
ones after they are archived.
(If you were preconfiguring a boatload of servers and the extra engineering
to do this preemptively looked like it might be cost effective even if only
a few servers in the field turned out to need it, then you'd want someone
doing this work who could figure out why it might be a good thing for
particular hardware.)
If you have different speeds of underlying media or i/o pathway separate
media this is also useful in ASM and is exploited by creating separate disk
groups.
Good luck,
mwf
_____
From: oracle-l-bounce@xxxxxxxxxxxxx [mailto:oracle-l-bounce@xxxxxxxxxxxxx]
On Behalf Of Vladimir Barac
Sent: Thursday, April 08, 2010 5:02 AM
To: oracle-l@xxxxxxxxxxxxx
Subject: Separating online redo logs from database files
Hello, listers
What are general pros and cons on separating redo logs from database files?
Both database and redologs are already residing on RAID1 volumes. So this
separation would only mean - move redo logs to separate mount point (RAID1,
still).
We have EMC consultant insisting that it is in line with best practices. Why
exactly is it a good thing, what "best practice" actually means - he can't
say.
Our systems are moderately used OLTP databases. I we had lots of waits on
redo log writes (and we don't), I would understand moving redo logs to flash
drives (for example). Or, as I have seen previously, database goes to RAID5
and redo logs to RAID1 - depends a lot on database usage, etc. etc.
Actual real life inputs are welcome.
Regards,
Vladimir Barac
Other related posts:
