Re: databases and RAID ...

From: "Fred Moyer" <fred(at)digicamp(dot)com>
To: "Peter Eisentraut" <peter_e(at)gmx(dot)net>, "Tom Lane" <tgl(at)sss(dot)pgh(dot)pa(dot)us>
Cc: "pgsql-admin" <pgsql-admin(at)postgresql(dot)org>
Subject: Re: databases and RAID ...
Date: 2002-05-25 19:45:12
Message-ID: ILEMKFGEMKDJNPOGGNCKAEAKDCAA.fred@digicamp.com
Views: Raw Message | Whole Thread | Download mbox | Resend email
Thread:
Lists: pgsql-admin

<adding my $0.02>

JBOD : just a bunch of disks, not raid in my opinion
Raid 0 : striping over disks, no redundancy, hence the Redundancy in
Redundant Arrays of Independent Disks is zero.
Raid 1 : Mirroring, full redundancy, Redundancy 1(00%)
Raid 4: see thread
Raid 5: see thread,Striping across multiple disks with parity. You have
one spare drive, 3 drives minimum, recommend more cause raid-5 is SLOW
Raid 10: A mirrored pair of striped arrays (1+0).

from working with various raid controllers (ide and scsi) here is my
feedback (please rebuke me if needed as I'm sure others on this list have
more experience).

Performance (fastest->slowest)
hardware raid -> software raid
raid 0 -> 10 -> 1 -> 5
Redundancy (most -> least)
hardware raid -> software raid
10, 1 -> 5 -> 0

some people say raid 5 is the most redundant but if you have over seven
disks your change of two drives failing becomes a statistical reality, hence
raid 5 is best suited for arrays of 5-8 drives. RAID 10 and 1 are both
mirrored but can be expensive. raid 0 is the fastest but don't count put
mission critical data on it - add another n disks and make it raid 10.

IDE vs SCSI:
I have run both controllers and have found both perform well. The stripe
size for raid 10 and 0 is important - make it as large as possible for
databases (256k on scsi and 1 MB on ide) since you want the disk heads to
read as much as possible before seeking again. for databases use scsi if
you can - use ide for streaming audio/video. databases performance relies
on being fast at random reads/writes and that's where scsi wins.

> Is there any rhyme or reason to the various "RAID n" designations?
> Or were they just invented on the spur of the moment?

The paper that introduced the term RAID used a numerical classification
for the various schemes. (So I guess the answer is yes.) The traditional
levels are:

0 Nonredundant
1 Mirrored
2 Memory-style ECC
3 Bit-interleaved parity
4 Block-interleaved parity
5 Block-interleaved distributed parity
[Hennessy & Patterson]

There are also other levels. One poster talked about RAID 10 which
appears to be a mirrored RAID 5.

--
Peter Eisentraut peter_e(at)gmx(dot)net

---------------------------(end of broadcast)---------------------------
TIP 1: subscribe and unsubscribe commands go to majordomo(at)postgresql(dot)org

In response to

Responses

Browse pgsql-admin by date

  From Date Subject
Next Message Andy Ruhl 2002-05-25 20:00:45 Re: databases and RAID ...
Previous Message Ragnar Kjørstad 2002-05-25 19:38:39 Re: databases and RAID ...