Re: RAID stripe size (was: 5.1 RAID5 write performance)

[Thor Lancelot Simon <tls%panix.com@localhost>]

On Mon, Jun 06, 2011 at 03:24:15PM +0200, Edgar Fu? wrote:
> > Ah, yes, the old 
> > rmwrmwrmwrmwrmwrmwrmwrmwrmwrmwrmwrmwrmwrmwrmwrmwrmwrmwrmwrmwrmwrmwrmwrmwrmwrmwrmwrmwrmwrmwrmwrmw
> >  cycle.  Gets me every time.
> OK, I've fixed that (before doing the tests I reported the last two days).
> 
> So, what's the advantage of a larger sectPerSU?
> It appears to me that the raidctl manpage should note that
> -- the stripe size should match fsbsize

Wait.  I didn't notice this until just now.  This is not right.

The filesystem block size (or, where this is not possible, the maximum
cluster size the filesystem will write) should be equal to sectPerSU
times the number of data (not parity) disks.  The sectPerSU value is
the *per-component* stripe size, not the total amount of data written
across all components in one stripe.

Thor