[nmglug] raid 10 vs 5

[Brendan]

On 09/09/2009 10:13 AM, Andres Paglayan wrote:
> quick question,
>
> If you have 4 x 1 TB disks, and they will be used for general purpose,
> (ie db and file storage)
>
> how would you arrange them?
>
> raid 5 with one hot-spare?
> or raid 1+0 (raid 10) stripping and mirroring?
>
> Both will provide 2 TB of available space,
> will the raid 10 be faster I/O?

Speaking for Linux mdadm software RAID, assuming no bus bottlenecks, etc:

10 will give you fast reads and writes.  Sustained reads won't be as 
fast as 5, but random reads will be faster.  Sustained writes will be 
slower than 5.  You will survive one disk failure.  You *might* survive 
two, but there's a only a 1 in 3 chance of that.  Linux has a native 10 
mode that is distinct from setting up a 0+1 or a 1+0.

5 will give you very fast reads, faster than RAID 10, even.  Some writes 
will be slower, but not markedly: There's some great tricks now that 
ameliorate the need to read all the disks to write to a single one. 
Sustained writes may be faster  You'll survive one disk failure.  If you 
have a hot spare, you'll probably survive two, assuming a second disk 
doesn't fail before the spare is put into play.

6 provides fast reads and sustained writes, similar to 5, but random 
writes will be slower.  You will survive two drive failures.  RAID6 will 
eat some extra CPU time, but it's largely inconsequential.

If data integrity and uptime is very high on your priority list, go with 
6.  If random IO performance is king, go with 10.  If you need 3TB of 
space, go with 5.  Whatever you do, buy server grade 24/7 rated drives 
and do regular backups.

-Brendan