With RAID-5, each write is, loosely speaking, broken down into five stripes
of data, and one stripe for parity, with each stripe going to a separate
disk (in RAID-5, parity is rotated among disks, where IIRC RAID-4 uses a
dedicated disk for parity). So if want your data to be stripe aligned, you
tell mkfs you've that each write is (n -1) stripes wide, where n is the
number of disks.
> -----Original Message-----
> From: Christian Guggenberger
> Sent: Thursday, December 11, 2003 6:15 PM
> To: Joshua Schmidlkofer
> Cc: linux-xfs@xxxxxxxxxxx
> Subject: Re: sunit, swidth question - mkfs.xfs tuning
> On Wed, 2003-12-10 at 00:23, Joshua Schmidlkofer wrote:
> > Howdy,
> > WOO HOOO!! On 2.4 inclusion, BTW. (good going!)
> > An old e-mail from Christian Guggenberger recommends swidth be (n-1)
> > disks for RAID5. Is that correct? I have a 66GB volume so I was
> > looking at:
> well, the (n-1)*sunit=swidth thing was suggested by Steve some long
> time (sorry I can't find the original message in the archives) - and
> that's what I'm using here on several Raid5 setups.
> > Mylex AcceleRAID 170, 6 disks in a RAID 5. (4GB system memory)
> > mkfs.xfs -d su=64k,sw=5 -l size=32m,sunit=64k -L "data" /dev/...
> it looks right for the data-section, but sunit=64k is way too
> large for
> the logsection.
> (the syntax is wrong, too - should be sunit=128, (or su=64k)
> as sunit is
> given in multplies of 512-byte block units)
> http://marc.theaimsgroup.com/?l=linux-xfs&m=104419883023845&w=2 for
> more info.