topic Re: MSA1500cs and internal array 5i Plus performance in Disk Enclosures

MSA1500cs and internal array 5i Plus performance

EdwardD — Mon, 31 Jan 2005 05:30:47 GMT

Hello,

There is the following system configuration:
1. Storage array: MSA1500cs with 2 MSA1000 controllers connected to one MSA30 SCSI enclosure with twelve 15k U320 disk drives in it.
2. At this time storage is connected through one MSA1000 controller only to one HP SAN Switch 2/8V (there are no more other systems connected to switch)
3. One HP Proliant DL585 server with Qlogic HBA connected to FC switch above. Server internal storage: four 15k U320 disk drives (same as in MSA30 enclosure) connected to Integrated Smart Array 5i Plus Controller on one channel (default server configuration).
4. Operating system: fresh install of Windows 2000 Advanced Server with SP4 installed by using SmartStart7.10 CD
5. additional software: HBA drivers installed from hp StorageWorks modular Smart array 1500cs Support Software CD Version 7.01-01 (latest from HP web)

Test environment and executed tasks:
1. Both Storages controllers (MSA1000 and 5i Plus) setting:
- cache distributing: read-50% ;write-50%
2. LUNs:
- external on MSA1500: one LUN 200GB on 12 disks on RAID0+1 (disk "D:" in Windows)
- internal 5i Plus: one LUN 70GB on 4 disks on RAID0+1 (2 partitions "C:" and "E:" in Windows)
3. Used performance test environment: Intel IOmeter
4. Tested disks performance: "D:" and "E:" with same IOmeter tests conditions
5. IOmeter performance reports created

Test results:
All disks performance tests shows, that internal disk E (placed on 4 disks in RAID1+0 on 5i Plus) is about 1,5 times faster than disk D (placed on 12 disks in RAID1+0 on MSA1000 controller)

Problem:
Could you explain those results and make your sugestions is it normal situation, because both raid controllers are Ultra3, both LUNs are on one channel RAID0+1, but MSA LUN is placed on more spindles.

There are helpful files (zipped to logs.zip) attached to mail where environment configuration, versions, test results could be found:
resultsmsa1500.csv - IOmeter results for MSA created LUN
results5iplus.csv - IOmeter results for 5iplus created LUN
switchcfg.txt - switch "supportshow" report
msa1500cs.txt - msa1500 config report
diag.xml - HP diagnostics report for server
iometertestcfg.icf - test condition for IOMeter

Re: MSA1500cs and internal array 5i Plus performance

John Kufrovich — Tue, 01 Feb 2005 11:34:14 GMT

Edward
When you ran the IOmeter, under the Disk Target tab. What did you set the "# outstanding I/O".

Re: MSA1500cs and internal array 5i Plus performance

Adrian Ogden — Thu, 03 Feb 2005 14:59:43 GMT

I've been experimenting with a MSA1000 and found a huge difference in sequential write performance in RAID0+1 and RAID0 configurations. I'm only getting 18MB/s in RAID0+1 against 80MB/s RAID 0

I'm still pondering why this should be so and will be trying some simpler tests.

Re: MSA1500cs and internal array 5i Plus performance

John Kufrovich — Thu, 03 Feb 2005 17:43:25 GMT

Adrian,
Need more information on your LUN configuration. How many spindles in the Raid 0 and how many in the Raid10.

Re: MSA1500cs and internal array 5i Plus performance

Adrian Ogden — Tue, 08 Feb 2005 04:43:09 GMT

I had two 15krpm 36Gb disks in both configs. Each drive was on a different channel e.g. at either end of the MSA1000 shelf (no external shelves).

Re: MSA1500cs and internal array 5i Plus performance

John Kufrovich — Tue, 08 Feb 2005 15:45:14 GMT

Adrian,
Your comparing a two disk RAID0 with two disk RAID 10.

What is your block size.

To be fair you will need to have four drives in your RAID 10. That way you will only see the double write overhead.

Re: MSA1500cs and internal array 5i Plus performance

Adrian Ogden — Tue, 08 Feb 2005 17:16:53 GMT

I'm using the max block size when setting up the array. I'll try your four disk config.

The results come from SiSoft Sandra.
Thanks

Re: MSA1500cs and internal array 5i Plus performance

Adrian Ogden — Wed, 09 Feb 2005 11:32:22 GMT

I got the same results with two disks on each channel set up as RAID10 (I call it 0+1). Seq Write of 23MB/s. Array was set up using highest chunk size, formatted default settings NTFS and tested with SISoft.

Re: MSA1500cs and internal array 5i Plus performance

John Kufrovich — Wed, 09 Feb 2005 12:04:33 GMT

I'm not familiar with the product.

There is a difference in RAID 1+0 and RAID 0+1. We, the MSA, uses RAID10.

If you would like, I can set my msa in the lab to duplicate your results with IOmeter.

Re: MSA1500cs and internal array 5i Plus performance

Adrian Ogden — Wed, 09 Feb 2005 16:38:35 GMT

That would be great.
Its a pecularity of my system that we are primarily interested in sequential I/O of quite large files (500MB) and would like to get to the bottom of this.

Thanks

Re: MSA1500cs and internal array 5i Plus performance

John Kufrovich — Wed, 09 Feb 2005 17:51:36 GMT

Adrian,

With IOmeter are you testing LUNS with the File system or RAW.

If you are testing with FS, IOMeter creates a large flat file. It will use all available space. IOmeter initially creates that file and if you kill the program. You could have a very small program that could easily fit into the cache of the MSA. If true, then what you are really seeing is large cache hits for the RAID 0.

Try using raw, just make it basic.

jk

Re: MSA1500cs and internal array 5i Plus performance

Adrian Ogden — Thu, 10 Feb 2005 11:47:45 GMT

Hi John

Might be a bit of confusion, I haven't used IOmeter yet. Sisoft Sandra seems to do the same sort of thing and I've been using that on a formatted NTFS partition.

I'll have a play with IOmeter.