[ale] Confusing RAID Performance

[Michael B. Trausch]

On Wed, 2011-02-02 at 16:23 -0500, Jeff Hubbs wrote:
> On 2/2/11 2:15 PM, scott wrote:
> > Remember that RAID6 is slower than RAID5.  RAID5 calculates the
> parity
> > once.  RAID6 does it twice.  This is to make sure that you have
> parity
> > protection incase you drop a drive. I would only recommend RAID6 on
> > large drives (1TB or larger).
> I sure wouldn't.  For >=~1TB drives, the probability of having an 
> unrecoverable read error among all the drives at recover time starts 
> becoming significant.  Sure, you can use it - as long as a restore
> from 
> tape, etc. is an acceptable fallback if you can't rebuild after a
> drive 
> replacement. 

I don't remember where I read this, but essentially the problem is that
the error rate over a unit of physical platter space has remained
relatively constant, and aren't any physically smaller.  However,
sectors have gotten a lot smaller (physically speaking), and so given
the same rate of error on a 60 GB drive and a 1 TB drive, the 60 GB
drive is more "trustworthy", in that you will lose a lower percentage of
it.

I have an array presently of five 750GB hard disks as RAID 6.  The array
was started with five refurbished drives, and once per calendar quarter
I replace one of those drives with a new drive (two more quarters, and
they will be all new drives, and I will stop replacing drives).  I am
doing this with the intention of (hopefully) staggering the drives
enough in age that the likelihood of catastrophic failure due to
triple-drive failure is near-zero.  However, I stand ready with a full
backup in order to restore its contents if ever that should happen.  I
do not believe that RAID is an excuse to have a poor or non-existent
backup plan; one should always be prepared to restore an array of any
size and configuration, IMHO.

RAID 6 is slower for writes, though being that it can sustain two
failures and still be operational, it is worth considering as long as
the array is battery-backed and as long as the system running the array
will shut down cleanly in the event of a power loss situation.  For
read-mostly loads, it provides lower overhead (unless the array is
degraded).

I didn't go with RAID 5 because I would hate to have the array fail a
drive and then attempt to recover and go completely dead because during
recovery, another failed drive is found.  I don't think it's terribly
likely, but I think it's far more likely than triple-failure for a small
number of drives.

	--- Mike
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 198 bytes
Desc: This is a digitally signed message part
Url : http://mail.ale.org/pipermail/ale/attachments/20110202/eaa009df/attachment.bin