[clue-tech] RAID no. of disks

[Angelo Bertolli]

mike havlicek wrote:
>
> --- On Wed, 4/8/09, Angelo Bertolli <angelo.bertolli at gmail.com> wrote:
>
>   
>> From: Angelo Bertolli <angelo.bertolli at gmail.com>
>> Subject: Re: [clue-tech] RAID no. of disks
>> To: "CLUE technical discussion" <clue-tech at cluedenver.org>
>> Date: Wednesday, April 8, 2009, 8:58 PM
>> Yeah, I've read it all before. 
>> With upwards of probably 2 Petabytes of 
>> data at this point, we're not going to spend the extra
>> money to have 
>> completely redundant data.  Everything is a tradeoff.
>>
>>     
>
> If you had 12 drives and wanted RAID 5 tolerance (not using any spares)
> what would be the comparison between say usable space and performance overhead in concatenating 2 packs of 6 disks, 3 packs of 4 disks, 4 packs of 3 disks, or one 12 pack to get one usable glob ?:) (Ooh and performance hit when you do start losing drives... or is it best not to glob it?
>
> -Mike
>   

Well for one, it depends on how it's implemented.  For example, we've 
had problems with disk failures when we've created md devices that 
consist of other md devices (i.e. RAID 1+0 over say 14 disks).  At least 
it was a problem to replace the drives without rebooting the machine, 
which is also a consideration for us.  But let's say you're ok with 
using 2 volumes and not creating a linear or striped array of 6 drives 
each.  If you're using RAID 5, you'd have 1 redundant drive in each 
array and only a total of 10 usable disks.  It's similar to the 12-disk 
RAID 6, with some minor differences.  3 packs of 4 disks would require 3 
redundant drives, and therefore only give you 9 disks of usable space... 
you get the idea.

When you start using specialized hardware, then things get even more 
interesting.  For example our best storage is actually technically RAID 
3 (with spares), but with special disk controllers, memory buffers, and 
a parallel architecture that make it perform at least a RAID 5 level, if 
not much better than you would expect.

Don't get me wrong, I think the advice Jack Parker mentioned is 
excellent for most people.  People shouldn't just choose RAID 5 because 
it's "what everyone else is using."  But in my experience RAID 5 & 6 
have only been a problem when there was a fundamental problem with the 
hardware (bad communication through the HBA, etc.)  and too many drives 
get dropped at once.  It hasn't been a problem simply from drive 
failures, and we do have a lot that fail.

Angelo