On Tue, Oct 13, 2009 at 02:56:17PM -0700, Michael Evans wrote:
> If you're only using 4 drives, Raid 10 (linux has a raid level that operates
> this way natively, but reshaping was not-yet supported when I last
> checked.)(or raid 0 across multiple raid 1 block devices) would yield
> superior performance across every dimension except fault tolerance.  That is
> the one advantage raid6 would have with that few devices, any 2 could fail
> and data would still be intact; while with the raid solution the right two
> drives failing could corrupt/destroy half the data.
> 
> Raid6 makes more sense with large quantities of devices (at least 6 IMO)
> when latency isn't as much of an issue as bandwidth.  In particular
> Databases and small file serving operations probably operate better off of
> raid10/1+0; while most operations on large files would benefit more from the
> raid 0 (across N-redundancy) devices performance of raid 5/6 for the typical
> case of fault free operations.

I understand this will be a bit of a religious issue, so I'll just start out
by stating things in a ridiculously dismissive tone:

Do not use RAID5 or RAID6 for anything at all. Ever. For any reason.

* The alternative, RAID1 or RAID10, is negligably more expensive, even for
  home use.
* RAID5/6 is slower, especially on writes.
* The larger the drive, the more likely that rebuilds will fail. My own
  recent experience with >1TB drives has been that a 4-drive array will
  rebuild itself into something usable a bit less than half the time.
* Rebuilds will silently spread any data corruption across all drives in the
  array (and the chances of such small amounts of silent data corruption are
  high, and increasing).

Reference: http://baarf.com
-- 
Robert Woodcock - [email protected]
perl -e '$a-=($_%4-2)*4/$_++while++$_<2e6;print"$a\n"'

Reply via email to