Re: Oarcle and RAID question

  • From: "Steve Perry" <sperry@xxxxxxxxxxx>
  • To: <DGoulet@xxxxxxxx>, <oracle-l@xxxxxxxxxxxxx>
  • Date: Thu, 11 Aug 2005 05:50:58 -0500

I'm just getting done running IO Meter against a CX700 on a Win2K host. We don't have *nix in-house and IO Meter kept choking on Win2003.
I ran it against RAW devices to eliminate the OS cache.
It may not be a perfect test, but here's what I tested:
Raid5 4+1 (all disks in single enclosure) - single 200 GB lun
Raid5 4+1 (spread over 2 enclosures) - single 200 GB lun
Raid5 8+1 (single enclosure) - single 200 GB lun
Raid1+0 ( 2 disks in 1 enc., 2 disks in another)
Raid5 4+1 2-200 GB Concatenated Metalun config (2 diff raid groups in 2 seperate enclosures ) - single 400 GB lun (recommended method)
Raid5 4+1 2-200 GB Striped Metalun config (2 diff raid groups in 2 seperate enclosures ) - single 400 GB lun (EMC doesn't recommend)

I used 1 Worker and started with 1 queued io per drive (ie. if it's a 4+1 config, I started with 4 and incremented by 4) and ran it up to 12 per drive. I also ran each test using 512Bytes, 4KB, 8KB, 16KB, 32KB and 64KB blocksizes.

the only io patterns I could run were 100% random read, 100% random write, 100% sequential read, 100% sequential write - I wanted a baseline. I know. not too many databases are doing 100% of one thing or another, but I had to start someplace.

The host is a 2 cpu, 2 gb ram, single emulex lp9002 hba (2 gb).

I haven't got to review the navi-analyzer files due to time. I'm trying to get it done before I leave. Friday is my last day before leaving storage and going back to being a DBA.

I can tell you the Clariions are normally setup for 8KB blocksizes and the results show that 8KB tests performed the best overall. Since the CX is in use, I couldn't make changes to set it different blocksizes so the deck may have been stacked.
We use Raid5 and on ours systems (Oracle, SAP - ISOIL, BW, HR, in-house) and things run just fine. The larger databases are only 2TB but I never saw IO problems against the CX. The bottlenecks we have are all cpu/memory issues.
We used R5 because mgt. only bought enough storage for what our growth had been and not what it turned out to be. EMC got behind the curtain and told them that they could get some extra space going to R5. In short, we never had an IO issue with R5 that couldn't be resolved, except with VMware ESX, but it wasn't with too much IO. we couldn't get it to push io fast enough compared to physical host, but I'll save that for another discussion.

We also have a DMX 1000 that I wanted to test. I'm going to see if I can get the replacement to run them.

Let me know if you're interested in the results. I don't know what the hardware diffs are between a 500 and 700, so you might want to check at EMC.


----- Original Message ----- From: "Goulet, Dick" <DGoulet@xxxxxxxx>
To: <oracle-l@xxxxxxxxxxxxx>
Sent: Wednesday, August 10, 2005 10:52 PM
Subject: Oarcle and RAID question

Yeah, Yeah I know it sounds like bug spray, but is there anyone out
there who has placed a database on an EMC CX500 SAN using raid 10?  If
so, what was the performance like?  Our EMC rep is trying to sell us on
the idea.

Dick Goulet
Senior Oracle DBA
Oracle Certified DBA


Other related posts: