RE: Re Raid 5+

Yechiel,

Just to add to Jared's note, RAID5 does a _lot_ more during a write. It has to perform what is known as the 'read-modify-write' cycle, where it has to first read the existing parity for the data that is being written. This is an I/O to _all_ the disks involved. It then has to calculate the parity which ties up CPU, and although it can be aurgued that this is done at the SAN level or controller level, the fact remains that additional work is involved. Finally, the newly calculated parity has to be written to all the disks involved. In other words, writes consume a lot more bandwidth on RAID5 as compared to RAID1 for the same amount of useful work.

This is not well documented and manufacturers hide this piece of work behind larger and larger caches. This is Ok for a small amount of writes, but following Tim's excellent explanation about pipes, you will face this soon if you flood your cache and your I/O subsystem. Needless to say, all manufacturer's claims are based on small writes that are not sustained and never stress the cache.

On the other hand, I have seen that if I can split the redo, SYSTEM, RBS and TEMP and other hot datafiles away onto RAID 1 disks, I _can_ live with a combination of RAID 1 and RAID 5 on a wellconfigured SAN. I have been successful with a few clients with this method when Cost has been a factor as far as disks/space go. I had to argue for this, but the client was able to see benefits that I was able to promise.

Shalom!
John Kanagaraj
Oracle Applications DBA
DBSoft Inc
(W): 408-970-7002

What would you see if you were allowed to look back at your life at the end of your journey in this earth?

• The opinions and statements above are entirely my own and not those of my employer or clients **

> -----Original Message-----
> From: Jared Still [mailto:jkstill_at_cybcon.com]
> Sent: Sunday, November 03, 2002 4:49 PM
> To: Multiple recipients of list ORACLE-L
> Subject: Re: Re Raid 5+
>
>
>
> Yechial,
>
> > You have 12 disks. In raid 0+1 you use striping across 6 volumes.
> > In raid 5 you strip across 11 disks, so you get almost
> double the work
>
> They didn't tell you everything they know about RAID 0+1.
>
> The disks may be striped into 6 logical disks, and they are
> written to as 6 logical disks, but when reading, all 12 disks
> are read from independentantly.
>
> So a RAID 5 is not actually doing more work a READ.
>
> But during a write, it's doing quite a bit more than RAID 0+1.
>
> > without returning and moving the r/w head on the same disk.
>
> Maybe if there is one user on the system, doing 1 read. Multiuser
> IO systems don't serialize the IO, at least not to the extent that
> the single user scenario requires. This is much like reorging a table
> so that all data is in a 'contiguous' space. It's an
> illusion, it doesn't
> really work that way.
>
> Jared
>
>
>
> On Sunday 03 November 2002 08:03, Yechiel Adar wrote:
> > Hello Ian
> >
> > I heard a lecture on raid 5 disks a few weeks ago.
> > The rational behind read 5 being faster then raid 0+1 is this:
> > You have 12 disks. In raid 0+1 you use striping across 6 volumes.
> > In raid 5 you strip across 11 disks, so you get almost
> double the work
> > without returning and moving the r/w head on the same disk.
> >
> > Yechiel Adar
> > Mehish
> > ----- Original Message -----
> > From: MacGregor, Ian A.
> > To: Multiple recipients of list ORACLE-L
> > Sent: Friday, November 01, 2002 2:48 AM
> > Subject: RE: Re Raid 5+
> >
> >
> > I cannot fathom Raid 5 being faster than Raid 1 tor
> writes. The real
> > question is, is it fast enough for your users. We happen
> to have a 650
> > terabyte database here. Even using Raid 5 disk storage would be
> > prohibitedly expensive. So we use a home-built hierarchal
> storage system
> > and store much of the data on Redwood tape drives. Users know that
> > requesting data from the Redwood drives will take some
> time. But they
> > were told to expect that. (The database is Objectivity not
> Oracle, and I
> > have nothing to do with it). The online data as opposed to
> the near-line
> > data is stored in Raid 5 arrays.
> >
> > What I don't know is what percentage of Oracle databases
> can run fine on
> > Raid 5 vs. Raid 1. It would not surprise me if the answer
> was well over
> > 50%
> >
> > Ian MacGregor
> > Stanford Linear Accelerator Center
> > ian_at_SLAC.Stanford.edu
> > -----Original Message-----
> > From: John Hallas [mailto:john.hallas_at_hcresources.co.uk]
> > Sent: Thursday, October 31, 2002 12:34 AM
> > To: Multiple recipients of list ORACLE-L
> > Subject: Re Raid 5+
> >
> >
> > Jared,
> >
> >
> >
> > We are certainly going to be performing extensive
> testing to ensure
> > performance of our applications under Raid5+ is acceptable.
> >
> >
> >
> > That means it is as good if not better than that
> experienced under
> > Raid1
> >
> >
> >
> > As I see it Oracle gain no benefit for stating that
> Raid5 should be
> > used if they did not believe that to be the case. If there
> was any doubt it
> > would be easier fro them to leave things as they were
> >
> >
> >
> >
> >
> > John
>
> ----------------------------------------
> Content-Type: text/html; charset="iso-8859-1"; name="Attachment: 1"
> Content-Transfer-Encoding: quoted-printable
> Content-Description:
> ----------------------------------------
> --
> Please see the official ORACLE-L FAQ: http://www.orafaq.com
> --
> Author: Jared Still
> INET: jkstill_at_cybcon.com
>
> Fat City Network Services -- 858-538-5051 http://www.fatcity.com
> San Diego, California -- Mailing list and web hosting services
> ---------------------------------------------------------------------
> To REMOVE yourself from this mailing list, send an E-Mail message
> to: ListGuru_at_fatcity.com (note EXACT spelling of 'ListGuru') and in
> the message BODY, include a line containing: UNSUB ORACLE-L
> (or the name of mailing list you want to be removed from). You may
> also send the HELP command for other information (like subscribing).
>

-- 
Please see the official ORACLE-L FAQ: http://www.orafaq.com
-- 
Author: John Kanagaraj
  INET: john.kanagaraj_at_hds.com

Fat City Network Services    -- 858-538-5051 http://www.fatcity.com
San Diego, California        -- Mailing list and web hosting services
---------------------------------------------------------------------
To REMOVE yourself from this mailing list, send an E-Mail message
to: ListGuru_at_fatcity.com (note EXACT spelling of 'ListGuru') and in
the message BODY, include a line containing: UNSUB ORACLE-L
(or the name of mailing list you want to be removed from).  You may
also send the HELP command for other information (like subscribing).