[Gllug] Large raid 5/5+0 setup

Kostas Georgiou k.georgiou at imperial.ac.uk
Thu Jun 25 21:24:14 UTC 2009


On Sat, Jun 20, 2009 at 07:42:59PM +0100, Nix wrote:

> On 19 Jun 2009, James Courtier-Dutton verbalised:
> > RAID 5 slows down systems that make small read/write transactions
> > because the RAID has to read the entire stripe for each transaction.
> 
> Is this really true these days? Modern disks can read (or write) your
> average RAID stripe (64K -- 4Mb) in essentially no time: i.e., it's
> vastly dominated by the seek time, and the drive electronics actually
> read multimegabyte hunks no matter what you ask for, in any case.

Raid5 never had to read an entire stripe, for random reads the stripe
size vs read size does matter but not because you have to read the
entire stripe [1] has a good description as to what is going on there.

For writes it is still very true, the read-modify-write cycle that you
get for partial parity writes is deadly to performance, you end up with
less IOPS than a even single disk.

For a mythtv box you only care about serial performance so none of the
above really matters.

[1] http://storageadvisors.adaptec.com/2006/06/05/picking-the-right-stripe-size
-- 
Gllug mailing list  -  Gllug at gllug.org.uk
http://lists.gllug.org.uk/mailman/listinfo/gllug




More information about the GLLUG mailing list