🚀 go-pugleaf

On Sat, 02 Oct 1999 11:50:21 +1200, Shaman <shaman@ihug.co.nz> caused to
appear as if it was written:

>Anyway back to the topic (RAid 1 vs 5; write performance  and overall
>performance for a file and print server with a hardware RAID card.)
>
>I was aware of the need to calculate parity during the write in RAID 5 but I
>thought the ability to write a small peice of the whole file to each disk (in
>RAID 5) would be more significant timewise.

Writing is always problematic, but you are correct in that *reading* data
written across multiple disks is pretty much always a win.

>Does RAID 5 ever get better than RAID 1,  as you add disks?

The short answer is... "define better"!

>ie RAID 1 vs   a  8  Disk RAID 5
>is RAID 1 still better?

As a rule of thumb, "powers of two plus one" work better, so 9 is a better
number than 8.

That said, if you are using a hardware controller, most of these things
support mirrored & striped and/or mirrored & concatenated "RAID1"
configurations.

So ultimately, one can say that, for a given usable disk capacity, a
configuration using RAID1 will pretty much always beat a configuration using
RAID5.

The justification for this is that disk array performance tends to boil down
to the being limited by the single-spindle performance, i.e. the performance
of the individual disk.

So, for an array providing a capacity equal to "n" disks, you'll have "n+1"
RAID5 disks capable of serving simultaneous read requests, and "2n" RAID1
(irrespective of whether they are mirrored & striped -- i.e. RAID0 & RAID1
-- or whether they are merely concatenated RAID1 drives).

So for your typical file server, where you'll have enough requests in the
queue to keep the thing busy, RAID1 will always win...

The "penalty" of RAID1 comes, as with RAID5, when you're writing: the fact
that the mirrored pair aren't spindle sync'ed means that one drive will
generally complete before the other, and the operation can't complete until
both are complete.  With RAID5, best case, you're dealing with a "full
stripe" write, which cannot complete until all drives complete...  So RAID1
is a win here, too.

>David

Malc.