Tracy R Reed wrote:
-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1
Tracy R Reed wrote:
A couple of very interesting studies have come out recently about the
reliability of hardware, specifically disks:
Someone from netapp has replied to the Google and CMU drive reliability
studies:
http://storagemojo.com/?p=388
Confirming what we suspected about SATA/SCSI/FC being the same
internally with the primary difference being the firmware. What I find
most interesting about NetApp's comments are how they rail against
RAID5. They say use only RAID1 or RAID6. This is in line with CMU's
suggestion that the odds of a double failure in a RAID5 is greater than
we think. I'm going to have to look into using Linux's RAID6 support.
Hopefully it is production ready. With drives so cheap and throughput
being such an issue going to more smaller/cheaper drives seems like a
better idea all the time.
You don't want a parity striping RAID (5 or 6) if you value write
performance.
NetApp can get away with it because they have nice battery-backed caches
that reorder and hide the whole read-modify-write cycle required with
parity striping. In addition, they have hardware accelerators to hide
the XOR required to generate the parity stripe.
If you're going software RAID and care even remotely about performance,
you want either RAID 1 or RAID 1+0, not RAID 5 or 6.
-a
--
[email protected]
http://www.kernel-panic.org/cgi-bin/mailman/listinfo/kplug-list
