Archive for February, 2010|Monthly archive page

HP SmartArray controllers penalizing “foreign” harddisks?

Another story from the hardware perspective.
As a backup server I’m running an old-ish DL380 G3. It doesn’t have too much power but can store some backups with 6 x 146GB SCSI drives in a Raid 5 configuration.
Some weeks ago one of the disks in the array died and needed replacement. As the machine is far out of guarantee I checked the costs for harddisks and went with a seagate drive in an hp tray, it has similar specs and hp seems to use seagate for at least some of its own drive offers.

Guess what happened to performance?
Rebuilding the array took like 20 hours. I wouldn’t have thought that it would be that bad but well, as long as it works … .
Afterwards I was shocked by backups taking more than half a day, where it usually takes 1-2 hours in the night. IOWait was around 80% by the time.
Hence I checked the io performance with iozone and got around 3MiB/s write speed. Doh!
I checked the results from the controller firmware … all disks are fine, no error, nothing, just that the last disk doesn’t have hp/compaq firmware.

That was with the internal HP SmartArray 5i with battery backed write cache. That controller supports only the older 160 MB/s SCSI standard. Well, old controllers are cheap. Let’s try a SmartArray 641 with BBWC which supports 320 MB/s.
Write speed went up to 5 MiB/s, great! 😉 Not really but better. Now I got another original hp harddisk and tried again … write speeds went to 22-41 MiB/s (without BBWC because the battery died)!

So, I’ve read and heard of mixing drives from different manufacturers in a RAID array is bad. But that bad? I mean, the drives have similar specs (SCSI U320, 10k, similar access times and all), what can go wrong with that? Still that could be the issue.
On the other hand, as the other hp harddisk made the difference I assume that the controllers are penalizing foreign harddisks. Either because of them not going through their own QA or to fuel their replacement parts business.

Go figure!

PS: one good thing about the SmartArray controllers I found in this process: they are really data compatible. The better SmartArray 641 recognized the array without problems. Turning off the SmartArray 5i brought the system suddenly back to life, no need to change or configure anything – beside the performance.