[H-GEN] Linux file server
Russell Stuart
russell at stuart.id.au
Thu Jul 1 21:38:45 EDT 2004
Tim Browne wrote:
> I have easy access to both Promise and Adpatec controllers.
> and
> Sony,LG,Mitsubishi DVDs.
My experience is that:
1. Linux software Raid is faster than hardware Raid,
providing you have the spare CPU cycles. This is
the consensus on the Dell PowerEdge list. Dell
seems reluctant to publicise this, probably because
it means they will sell less high-end raid cards.
Be that as it may, when pressed, the verdict of the
Dell engineers seems to be unanimous. Qualification:
I have not done the tests myself.
2. Software Raid is more reliable that hardware Raid. This
is primarily because of bugs in Raid cards and firmware
bugs/incompatibilities in the SCSI disk drives. The
main trouble area seems to be subtle incompatibilities
between drives and card giving rise to timing and protocol
errors. The errors have, in my case anyway, invariably
lead catastrophic loss of all data in disk array. This
may seem odd, but it arises because there are several
manufactures of Raid cards, and there aren't that many
of each type sold. The Linux Raid software gets used
by a lot of people. The HDD drivers it depends on gets
thrashed every Linux box.
3. Raid is less reliable than a single disk spindle. In
other words, if all you data fits on a single drive, then
you will better off reliability wise to just one drive
instead of Raid. I do not know if this holds true for
software Raid - I haven't used it enough. It is definitely
true for hardware Raid. Every Raid array I have admined
has failed (as in required me to go to a backup). They have
always been on high-end servers, of course, and their
failure has caused a great deal of pain. Sometimes the
failure is due to hardware (cards / cables / drives / power
supplies), but more often due to aforementioned bugs.
Note the "single spindle" qualification. If you have to go
to multiple spindles for size or speed reasons, then
the outcomes start to favour Raid. If you have a large
number of spindles you won't be able to survive without
Raid - I have seen sites that average one drive failure
per week, because of the number of drives they use.
Without Raid the downtime would be simply unacceptable.
It might seem this does should not hold for Raid 1 (mirroring).
It does. Another way to use the second drive is as a hot
spare - doing an image backup each night after checking
that all you data looks consistent. This is more reliable
because data is damaged more often by applications (bugs
corrupting files) and users (eg deleting the wrong file) than
by hardware failures. If you are mirroring the damage is
reflected on all drives immediately. If you do a nightly
backup you have a reasonable chance of recovering the data
while the system is online, without messing around with backup
media.
The fallacy younger admins fall into (well I did and I assume
others do as well) is that since big sites use Raid to
get good reliability, their small site should also use Raid
for the same reason. While it seems reasonable, it is wrong.
The other thing Raid can give you is speed, but again if you
are a small site with a high read/write ratio spending the
money on RAM will give you more speed than Raid - without
the reliability concerns.
--
Russell
More information about the General
mailing list