[zfs-discuss] Are ST4000DM000 drives good for RAID-Z?

Michael Kjörling michael at kjorling.se
Tue Aug 27 13:07:36 EDT 2013

On 27 Aug 2013 08:16 -0700, from linux at nashira.ca (Bryn Hughes):
> Enterprise-grade drives usually are rated at 1 read error per 10^15 bits
> read btw, that works out to a much "safer" number (around 7200TB).  You
> need to check the specs - "value" SATA disks for example are often still
> rated at 1 in 10^14 whereas SAS drives are much more likely to be 1 in
> 10^15.  It is a whole order of magnitude different.
> This is why we all talk about the risks of using the desktop drives. 

And here's something actually quantifiable. Thank you. Yes, 10^14 bits
read before an unrecoverable read error was high when a drive was on
the order of 10^10 bits (~1 GB), but it's a radically less impressive
figure with drive sizes on the order of 10^13 bits (~1 TB) and beyond.
I can certainly see the reasoning there.

I've been going back and forth in my mind between some different
options and also took a few minutes at work today to talk it over some
with one of the sysadmins there (we don't run exactly _huge_ data
stores, but we do handle a fair amount of customer data as well as
corporate-internal data and nobody likes being on call when the
database server starts complaining about table corruption in the
middle of the night), and it looks like I'll be putting something
together with enterprise grade drives and continue making regular
backups as a safety net.

_It almost certainly won't be perfect_ (nothing ever is), but together
with regular scrubbing (I see the Best Practices Guide suggests¹ doing
so monthly for such drives) it should help me sleep well at night. At
the very least, I'll find out quickly about any data corruption on
disk. Also, in this case, perfect is probably the worst enemy of good
enough as my current ext3 setup won't do _anything_ to even notify me
of a read error in file contents unless it actually shows up on the
screen or causes an application-level checksum mismatch of some kind,
at which point the error may have been laying dormant on disk for a
long time!

¹ http://www.solarisinternals.com/wiki/index.php/ZFS_Best_Practices_Guide#Storage_Pools

Michael Kjörling • http://michael.kjorling.semichael at kjorling.se
                “People who think they know everything really annoy
                those of us who know we don’t.” (Bjarne Stroustrup)

To unsubscribe from this group and stop receiving emails from it, send an email to zfs-discuss+unsubscribe at zfsonlinux.org.

More information about the zfs-discuss mailing list