[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: RAID Suggestion for webserver



On Sun, Feb 10, 2002 at 02:14:32PM -0800, Lev Lvovsky wrote:
> err, unless I'm mistaken:
> 
> RAID10:
> mirrored system = total/2 = 80
> striped system = single * 2 = 80
> 
> combined that gives you a mirrored & striped "drive".
> 
> this gives you 80GB total

correct.  raid10 gives "n/2" capacity (whre n=number of drives).

> RAID5:
> 
> the "spare" is distributed over all of the disks...in the case of 4 disks,
> one of the disks can be thought of as a spare, which leaves you with 3 *
> 40 = 120 GB.

nope.

the spare drive is a dedicated spare drive, it is NOT distributed over
all the disks.  if one of the drives dies, then the system automatically
rebuilds the array using the hot-spare.  you're thinking of the parity
data - in raid5, the parity IS distributed over all disks.

raid5 gives "n-1" capacity, or "n-2" capacity when used with a hot-spare
(or "n-3" if given two hot-spares and so on).

so a 4 x 40GB drive raid-5 array would give either 120GB (no hot-spare)
or 80GB (1 hot-spare)


> > The question becomes... which provides more performance and is more
> > reliable?
> 
> the raid10 from the looks of it...

raid 10 is slightly more reliable in theory (but only if you're lucky).
in practice...who knows? probably about the same.

raid5 is faster than raid10, which should be obvious when you think
about it.  raid0 (stripe) is fast, but raid1 (mirror) is very slow...it
is certainly much slower to write the same data to two drives than it is
to write to the data to one drive plus a parity calc & write.  with
raid10, you end up with a striped (double-speed) mirror (half-speed),
for no net speed gain over a single drive.  it doesn't get any faster if
you add more disks, either.

raid5 on the other hand stripes data and parity over all disks, so the
more disks in the array the faster it gets.


i did some benchmarks last year with an 8 drive scsi hardware raid
system to determine what configuration would provide the best
performance.   i tested different raid configurations, different
filesystems (xfs, reiserfs, ext2), and different usage patterns
(simulations of file-server, mail-server, database-server etc type
loads)

like you, i expected raid10 to be the fastest because that's what most
of the info on the net says.  nope.  i was surprised to find that raid5
was significantly faster than raid10.


FWIW, i don't think it's possible to generalise about raid performance
and be at all accurate.  it depends very much on whether you have
hardware or software raid AND on whether your raid controller has a
large *non-volatile* write-cache.

without the non-volatile write-cache, raid5 is very slow.  with it, it
is very fast.


craig

-- 
craig sanders <cas@taz.net.au>

Fabricati Diem, PVNC.
 -- motto of the Ankh-Morpork City Watch



Reply to: