Comments on: Effect of drive count on RAID-5

Comment on Effect of drive count on RAID-5 by: Tom

Wed, 18 Jul 2007 13:15:47 +0000

Maobo, great point about the power consumption. In fact, I suppose a post about power consumption per IOPS for various sized drives (3.5″ and 2.5″) would be mildly interesting. I might throw one together over the next few weeks.

As far as the reference to a 10X reliability decrease, I think that concept has become generally accepted as fact by the industry and I’m not sure I’ve actually got a document that has the science behind it. In fact I’m positive the 10X is just a Wild Ass Guess. It’s just meant to reflect an increased chance of failure due to correlated causes. I’m sure the actual number will vary from 2X to 100X depending on the cause of the failure.

Comment on Effect of drive count on RAID-5 by: Tom

Wed, 18 Jul 2007 13:07:10 +0000

Hi, Dean. Thanks for the great feedback.

As far as commonly accepted practice, I’m not sure I have a document that clearly outlines an industry consensus. I suppose your best bet is to “simply” read lots of white papers from other RAID controller vendors. That’s going to be tedious and probably won’t be as enlightening as you’d hope.

But I’d be happy to help supply other data if you’re second-guessed by the PHB. Good luck!

Comment on Effect of drive count on RAID-5 by: Tom

Wed, 18 Jul 2007 12:30:35 +0000

Barry, you’re right. My equation for rebuild time was a little simplistic. I’m sure most RAID controllers read the data in parallel. So the data will come from disk in parallel, and with SAS or SATA the data will flow across the bus in parallel. The only time the data is serialized is into controller memory which should have a high enough bandwidth to not be an issue. There’s also a potental issue with the XOR source count being less than the number of drives being read, in which case multiple XOR passes will be required.

I’ll go back and tweak my words in the original post. Thanks for catching that, Barry.

Comment on Effect of drive count on RAID-5 by: the storage anarchist

Tue, 17 Jul 2007 23:03:17 +0000

I think your rebuild time algorithm overlooks the fact that you can probably read the data across multiple disks faster by posting read requests to all of them at once. This way you can minimize the overhead of rotational latency, and perhaps even some of the seek time, that you otherwise encounter reading a single drive sequentially. In fact, with sufficient caching and pre-fetch, it’s possible to cut rebuild times by a significant fraction of the algorithm you propose.

But as you say, YMMV.

Comment on Effect of drive count on RAID-5 by: maobo

Fri, 13 Jul 2007 00:24:02 +0000

Another problem I think is the energy consumption. For one R/W it will generate more disks operations. Thus increased the overall power consume.

TT: Some folks like to correlate multiple failures, meaning that if one disk fails then there is a higher chance (such as 10X) that a second disk will fail ….

Can you give me the paper title which does the insight. Thank you!