We just installed a MD3000 & MD1000. We ran some simple performance test
using XDD and iozone. We tested various raid group at the device and file
system level. We are seen some very poor performance.
Single Disk Raid0 sequential Writes= 20MB/sec
4+1 and 7+1 Disks Raid5 Sequential writes = 100MB/sec
Is this performance typical? Any tuning suggestions?
I've recently installed a MD3000 with a red hat 5.3 host on a SAS 5/e and am experiencing similar performance figures also using various raid types and file systems.
We are running an MD3000 with dual SAS controllers (non-redundant with 3 hosts connected to non-shared Luns). It is chained to an MD1000. Both are populated with 1TB SATA (7200) drives. Running IOMETER on our Windows 2003 x64 servers we've seen:
6+1 Raid5 w/256K Segments (non-aligned partition blocked at 4k) :
We are somewhat baffled at the disparity between the single stream tests .... We would have thought that the aligned w/larger (Windows) blocksize would yield better results --- even on a single stream (It certainly does on multiple streams). The 6+1 Raid is entirely on the MD3000, while our 7+1 Raids all have at least 1 drive on the MD1000 extension. We were wondering if there might be a troublesome disk (disks?) on the MD1000, but everything appears Optimal. However, we ran a System State report via the CLI interface and are seeing a LARGE number of (Unit ATTN) I/O errors attributed to the (MD1000) enclosure itself :
Client: 02.71.G6.07 built at Wed Mar 12 11:13:22 CST 2008
There is much more to the report that I won't post due to its size. We're about to open an incident with Dell, and are wondering if anyone else is seeing numbers like this?
Are you sure all these FW's are up to date, check the EMM's on the 1000 expansion chasis as well?
We use MD3000i's and I was not really able to extrapolate to much to your running setup here.
We span stuff all over the MD3000i's with MD1000's attached. So no specific issues with that as far as I know. We have found if things get flakey? It is not a bad idea to reset the power supplies and in our case the controllers in these chasis. They get a little loose over time, moving them around, ect..ect...
Both Controllers on the MD3000 are at the same FW level ( 06.70.15.60 ) .
The MD1000 shows (for both EMMs): EMM card status: Optimal Firmware version: A.04
We are running a similar setup at another site and they too are seeing the high UnitAttn error counts. I believe that they have already update the FW on their MD3000 to the latest level, so I don't think that's going to change much. I'm going to call it in to Support and see what they find.
cbryce
2 Posts
292
0
Posted June 3rd, 2009 05:00
I've recently installed a MD3000 with a red hat 5.3 host on a SAS 5/e and am experiencing similar performance figures also using various raid types and file systems.
Did you manage to improve on this?