Re: New Disk Setup

From: Mark D Powell <Mark.Powell_at_eds.com>
Date: Thu, 28 May 2009 13:15:53 -0700 (PDT)
Message-ID: <cfcbd11e-056b-4894-ad6d-cd3feca71b8a_at_n4g2000vba.googlegroups.com>



On May 28, 1:02 pm, joel garry <joel-ga..._at_home.com> wrote:
> On May 28, 6:19 am, billshatne..._at_googlemail.com wrote:
>
>
>
>
>
> > On May 28, 12:55 pm, johnbhur..._at_sbcglobal.net wrote:
>
> > > On May 28, 5:34 am, billshatne..._at_googlemail.com wrote:
>
> > > > I've got to shift our OLTP production database onto a new server which
> > > > has 8 disks.
>
> > > > The layout I've got at the moment is the following.
>
> > > > Please feel free to flame me if you like but I was wondering if I'm
> > > > being overly cautious with mirroring and should use striping/
> > > > mirroring.
>
> > > > Any suggestions for modifications gratefully received.
>
> > > > 1 + 2 RAID-1
> > > > OS
>
> > > > 3 + 4 RAID-1
> > > > Datafiles
> > > > Control file
>
> > > > 5 + 6 RAID-1
> > > > Archived Redo Logs
> > > > Control file
>
> > > > 7 + 8 RAID-1
> > > > Oracle Software
> > > > Control file
> > > > Backups
>
> > > Well if you have a system with modest IO requirements almost anything
> > > will work.
>
> > > You don't give any details about what kind of server or disks or
> > > operating system or oracle version or if you are going to use ASM
> > > etc.  No information on what "kind" of backups either.  Does your os
> > > support asynchronous writes?  Are you planning on using RAW or using a
> > > cooked file system?
>
> > > With 8 disks available you are not going to get to get a lot of
> > > choices or performance.
>
> > > How about RAID 1 for disks 1/2 and put the OS + oracle software over
> > > there.
>
> > > Disks 7/8 for archive logs and backups in RAID 1.  You really don't
> > > want rman disk based backups writing to the same disks that the
> > > operating system is on ... that's guaranteed to cripple performance.
>
> > > That give you a whole 4 disks to work from eh?
>
> > > For many OLTP systems especially if they are commit happy the IO
> > > performance of the oracle online logs ( which you appear to have left
> > > out of your original description ) is critical.  How are you planning
> > > on meeting those requirements?- Hide quoted text -
>
> > > - Show quoted text -
>
> > Thanks for that. We're on Red Hat AS 4.7. My unix colleague is looking
> > to use logical volumes.
> > Disks are 2x36Gb, 6x146Gb.
>
> > Couldn't tell you about the server itself or more about the disks,
> > unfortunately. Oracle version is 10.2.0.4. We're not using ASM.
>
> > Backups will be hot backup and RMAN if we have the space.
>
> > Sorry, the online redo logs will be on a disk separate to the archived
> > redo logs. Currently they're 25Mb each.
>
> > The 4 disk solution sounds a lot better.
>
> I must say, even though my system may be way different than yours, my
> redo is 1.5G each (x3 or 4 x several databases).
>
> Also, I've found the RMAN compressed backup gives me around 10:1
> compression.
>
> Even though I' a BAARF member, I've found RAID-5 does work for normal
> situations (though it degrades rapidly for slightly-out-of-normal
> situations, and fails miserably if you lose 2 physical disks - and
> groups of disks installed at the same time, going through the same
> environmental stresses, often fail at the same time, MTBF ratings be
> damned).
>
> Google the SAME paper if you haven't seen it.
>
> jg
> --
> _at_home.com is bogus.
> "I couldn't help but notice you are glowing." - (more than one) bum to
> me.  http://news.bbc.co.uk/2/hi/science/nature/8070252.stm- Hide quoted text -
>
> - Show quoted text -

"I've found RAID-5 does work for normal situations"

I agree, while RAID-5 is not the ideal disk setup for a heavy hitter system it is often more than good enough especially since RAID devices often have significant write cache as part of their set up.

When the RAID-5 is done only via the OS with no disk cache then it is easier to run into write performance related issues but read is generally still decent thinks in large part to the stripping.

The fact that RAID-5 only protects you from the loss of a single disk in the stripe is one reason I suggested two 4 disk stripes rather than just a single 8 disk stripe but I have seen places use stripes with over 20 disks in them. (If a single disk is likely to fail once in 300 years how many disk failures would I expect if I had 300 disks?)

HTH -- Mark D Powell -- Received on Thu May 28 2009 - 15:15:53 CDT

Original text of this message