[clug] forget RAID?
chris at kororaa.org
Wed Feb 21 05:56:47 GMT 2007
> So (not being overly knowledgeable in this are) what does this mean in
> practical terms?
I'm no expert but in my opinion, the thing to realise is that hard
drives die. They die quicker and more easily when they are not operating
within their spec in regards to things like temperature. They are also
unreliable if they aren't connected to a decent power source. So if you
have two hard drives in a computer and one dies, then you have a high
possibility that the other drive, operating in the same conditions, will
also die at some stage. But in reality all hard drives die. All, every
one, every single one ;)
I agree that raid5 is not really a great level of protection. If you
lose one hard drive, then you cannot afford for another to die, yet to
re-build the array you need to 'stress' the rest of the drives to
re-build the newly inserted drive from parity information. That means
you have a higher chance that one more hard drive will die and if it
does then you lose all your data. Of course you have the same problem in
a 2 x drive raid1 system, but in a 4 x drive raid5 you triple your
chances of losing your data in a rebuild because any one of the
remaining three could die.
Personally for a database server with valuable data I choose raid 10,
which is similar to raid1 and results in a 50% space loss but means I
get a little performance boost. In a 4 x raid 10 array I can afford for
2 drives to die before I lose my data. I have also found drives can just
drop out of a raid array where there are parity calculations involved,
like raid 5 or raid 6. Not sure why, maybe that drive didn't write the
parity info properly or something.
And remember it's not just about hard drives dying, it's also about hard
drive corruptions, bad sectors and things that like that which affect
the integrity of your array.
> I currently have two servers on site - one backup and one web server.
> Both have 2x200GB HDDs in software RAID1.
> What sort of system should I setup to apply "cluster storage file
> replication" or is RAID1 the most sensible option for my environment?
Well raid 1 is essentially one hard drive and an instant backup.
Depending on your server you might only be able to fit two hard drives.
Raid 1 is the best option then. Why would you get rid of raid1 and have
a single drive running and then backing up to another? Just use raid1
and let it do it for you, and if one drive dies you have no down time.
> Is it as simple as making a backup server with 3HDDs, and setting up
> some sort of automatic copying between them, rather than using RAID1?
> - I'm not sure how this would help the problem noted in the article.
> Or would I create two partitions on each drive, and copy hda1 to hdb2,
> and hda2 to hdb1, for instance?
As above I don't see a problem with the use of raid1 for this. There is
no parity calculation involved, it's just the one set of data on a hard
drive written to another drive. So if you use a backup server with 3
hard drives, then I would run it in a raid1 array. This means you have 2
hard drives in raid 1 and you could then use the other hard drive as a
backup of the raid1 array. If you had four drives you could make two
separate raid 1 arrays and synchronise data from one to the other which
means you have the same data on all 4 drives.
Not sure if that helps, or is entirely right, but there you go :)
More information about the linux