ZFS RAIDZp read performance: 1x or (n-p)x ? (Was: Re: [zfs-discuss] Re: Suggestions for heavy activity build)

Durval Menezes durval.menezes at gmail.com
Mon Aug 5 12:11:34 EDT 2013


Hi everybody (but specially Gordan and Dan),

On Mon, Aug 5, 2013 at 10:55 AM, Gordan Bobic <gordan.bobic at gmail.com>wrote:

> Yup, pretty much what I figured. On reads 4-disk RAIDZ2 will have 2 disks
> worth of IOPS.
>

Are we sure about this? The "conventional wisdom" I've read here on this
list<https://groups.google.com/a/zfsonlinux.org/d/msg/zfs-discuss/VBJnKSW9CRk/28Ab1GNmO1IJ>and
elsewhere<http://constantin.glez.de/blog/2010/06/closer-look-zfs-vdevs-and-performance#raidzperformance>is
that RAIDZp random read performance is 1x the speed of a single disk.

It seems the disagreement comes from whether ZFS needs or not to read in
the whole stripe during normal operation (ie, no errors): this blog
post<http://utcc.utoronto.ca/~cks/space/blog/solaris/ZFSRaidzReadPerformance>,
for example, specifically says:

"in ZFS, reads always involve all disks, because ZFS always must verify the
data block's checksum, which requires reading all of the data block, which
is spread across all of the drive"


So, are we really sure that ZFS doesn't need to read from all disks in
order to verify the block checksum?


> On writes it will be 1 disk worth of IOPS.
>

On this everyone seems to agree.

Cheers,
-- 
  Durval.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://list.zfsonlinux.org/pipermail/zfs-discuss/attachments/20130805/5f795fa1/attachment.html>


More information about the zfs-discuss mailing list