[Gluster-users] RAID options for Gluster

[Brian Candler]

On Thu, Jun 14, 2012 at 01:21:29PM +0000, Fernando Frediani (Qube) wrote:
>    Well, as far as I know the amount of IOPS you can get from a RAID 5/6
>    is the same that you get from a single disk. The write can not be
>    acknowledged until it is written to all the data and parity disks.

Nope.

Worst case: a single block write will cause reads on two (RAID5) or three
(RAID6) disks, followed by writebacks to the same two or three disks.

Now, if you have an array with say 12 disks, then clearly only 1/6 or 1/4 of
the possible disk I/O capacity is being used; other writes can happen in
parallel, so the total array throughput is definitely higher than a single
disk.

However the *latency* for individual writes will be much higher than a
single disk, so for a single thread doing sequential writes to the array, it
will perform very badly.

Best case: a large write which fills the whole stripe will write different
data to all the disks simultaneously.  An array with N data disks will have
write bandwidth of N x a single disk.  Controllers with battery-backed cache
attempt to aggregate writes in this way.

Regards,

Brian.