RAID 5 and Performance Degradation Concerns

Printable View

10-03-2003, 05:23 AM
hacketta1

RAID 5 and Performance Degradation Concerns

Dear All

My manager (not a DBA) has put together a spec for a new disk array (Hewlett Packard XP 1024) on which we'll be running our 600GB OLTP database (64bit 8i and HP-UX 11i) and has asked for my opinions. My most immediate concern, and one which I've mentioned to him, is that it only has RAID level 1 or 5. Now we definitely need some hardware level striping so RAID 5 is our only option. I'm no storage expert, but understand that Oracle and RAID 5 is generally percieved to be a bad idea on a write heavy application due to the maintenance of parity information. He's responded that HP claim that on this array RAID 5 performs just as well as RAID 0+1 (my preferred option, we currently have an EMC Symmetrix and utilise this). I'm meeting some HP reps on Monday and want to grill them on this, but I was wondering if anybody out there has any practical experience with Oracle and this product, or any opinions on HPs claims.

Any guidance would be much appreciated.

Thanks

Austin
10-03-2003, 08:44 AM
robertbalmer

I will quote from the Oracle 9i bible(Performance Tuning Tips and Techniques)....p.65

"Raid 1+0 This level provides mirrored disks and striping. This incorporates the advantage of the first two RAID levels by providing the disk I/O benifit of RAID0 to the redundancy provided by RAID1.For high read/write environments such as OLAP,where sporadic access to data is the norm, this RAID level is recommended".

and

"RAID5 ...It is a low-cost solution that is inefficient for write-intensive applications."

--If you can get the book over the weekend read chapter 3.--
10-03-2003, 08:56 AM
tamilselvan

No matter what HP says about RAID 5, it will hurt perf for an OLTP system.
10-03-2003, 09:00 AM
marist89

Re: RAID 5 and Performance Degradation Concerns

Quote:

Originally posted by hacketta1
He's responded that HP claim that on this array RAID 5 performs just as well as RAID 0+1 (my preferred option, we currently have an EMC Symmetrix and utilise this).

Tell them to put up or shut up. Have them give you two boxes to evaluate, one configured with RAID 1+0 and the other with RAID 5 and run your application on them. If there's no difference, go with the RAID 5.
10-03-2003, 09:08 AM
tamilselvan

One more test you can do.

When DB is running, ask the HP guy to pull out 2 disks randommly selected in the striped set, and now you see your DB is up and running. Or the disk array itself is crashed.

Tamil
10-03-2003, 09:44 AM
hacketta1

Thanks for your responses. I like the idea of doing some tests of my own; not sure if they'll agree though. One thing I'm thinking is that the array will have at a minimum an 8GB cache, so if this could be configured as write-only it may go someway to circumventing the RAID-5 write penalty.

Something else that's vexing me is that it apparantly this thing has a minimum stripe size of 9GB. Again, I'm no expert in these things but I can't really see how this would help eliminate hot-spots.
10-03-2003, 10:37 AM
marist89

Quote:

Originally posted by hacketta1
I like the idea of doing some tests of my own; not sure if they'll agree though.

We eval stuff all the time. If they don't like the idea, tell them you are looking at adding some EMC arrays...
10-06-2003, 01:55 PM
marist89

http://www.miracleas.dk/BAARF/1.Mill...1.03-RAID5.pdf