RE: Losing data from Postgres

IMHO the biggest problem with RAID 5 is the performance. Under
normal conditions RAID 0+1 have double read performance (since each mirror
can be accessed indepently) and in case of an error things get really bad
for RAID 5.

If a disc fails iin a RAID 5 every read to the failed disc (or disc-stripe), requires read
from all other discs to reconstruct data, which decreases the bandwidth to the
discsystem dramatically.
Since most systems do most reads, this is really serious.

On a RAID 0+1 (some times called RAID 10) only the disc that has lost it mirror
decreases its bandwidth to "normal" single disc speed. All others keep running
a double speed. And actually write performance increases since the failed
mirror can be updated.

Another point is reliability. If disc number 2 fails in a RIAD 5 we goes down,
if we eg. have a 2 x 4 RAID 10  system, with a single disc failiur, the chances
that the second disc fail will cause break down id on 1/7, since 3/4 of the
disc system is still redundant.

Double failiur is not happening often, but has happend if all discs is installed
at the same time.

If I should set up an important server I will go for RAID 10, mostly because
on a heavy loaded server, the disc error bandwidth decrease in RAID 5 is the same as
not availably. 

/Thomas

-----Original Message-----
From:   Hossein S. Zadeh [SMTP:hossein@hossein.bf.rmit.edu.au]
Sent:   Thursday, November 16, 2000 7:17 AM
To:     admin
Subject:        Re: [ADMIN] Losing data from Postgres

On Wed, 15 Nov 2000, Alfred Perlstein wrote:

> * Serge Canizares <serge@ephilosopher.com> [001115 08:23] wrote:
> >
> > Of course, if someone sees a reason that RAID 5 would be better than RAID 1+0,
> > I'd appreciate an explanation!
>
> Cost. :)
>

A little bit more explanation: :-)

RAID 1+0 gives you only half of the installed space but gives you (n/2)
times speed of individual disks. For example for 4 hard disks 1G each, you
get only 2G of space but double the speed of individual disks.

RAID 5 gives you space equal to (n-1) times individual disks. This is far
better than RAID 1+0. For example for 4 hard disks 1G each, you get 3G
space (this is only 25% waste compared to 50% for RAID 1+0). As you add
hard disks to the array, the 25% ratio of RAID 5 get lower and lower, but
that of RAID 1+0 stays at 50%. For 10 hard disks for example, the ratio
gets down to 10%.
Speed of RAID 5 however is very much dependant of a few factors: speed of
the controller (or CPU speed in case of software RAID), type of data, and
how the array is setup (how many blocks of data per strip, etc.). In
theory, it can exceed speed of RAID 1+0, but I have never seen it in real
life (but it does approach that of RAID 1+0 if you spend $$$ on the
controller or CPU).


Hope it helps,
Hossein