zfs discuss - Nov 2009 - Disk I/O in RAID-Z as new disks are added/removed

Hello,

As I understand it, in a traditional RAID 5 setup adding new disks to the pool
provides more overall I/O as the load is spread out across multiple disks.

What exactly is this relationship in a RAID-Z setup? What should one expect in
terms of overall I/O performance as disks are added and/or removed? I understand
that the checksum data is distributed across all disks, unlike a traditional
RAID setup. Does this carry a significant performance hit? As far as
distribution of parity data/leveling, is there some sort of buffer or some sort
of means to give new writes priority over this distribution of data?

Any general information you can provide me as far as the theoretical concepts
behind increasing I/O by adding disks to a RAID-Z pool would be appreciated as I
assess this technology :)

Thanks in advance!
-- 
This message posted from opensolaris.org

Anybody?

I would truly appreciate some general, if not definite insight as to what one
can expect in terms of I/O performance after adding new disks to ZFS pools.
-- 
This message posted from opensolaris.org

On Sun, Nov 15, 2009 at 2:57 AM, besson3c <joe at netmusician.org> wrote:
> Anybody?
>
> I would truly appreciate some general, if not definite insight as to what
> one can expect in terms of I/O performance after adding new disks to ZFS
> pools.
>
>
>
I''m guessing you didn''t get a response because the first
result on google
should have the answer you''re looking for.  In any case, if memory
serves
correctly, Jeff''s blog should have all the info you need:
http://blogs.sun.com/bonwick/entry/raid_z


--Tim
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://mail.opensolaris.org/pipermail/zfs-discuss/attachments/20091115/44160d44/attachment.html>

On Sun, Nov 8, 2009 at 12:05 AM, besson3c <joe at netmusician.org>
wrote:
> Any general information you can provide me as far as the theoretical
concepts behind increasing I/O by adding disks to a RAID-Z pool would be
appreciated as I assess this technology :)
As with RAID5 and RAID6, a wider stripe will improve read performance
for raidz. Writes will generally be limited to the throughput of your
slowest device. On average, writes will still be faster than than
RAID5/6, since there is no read / re-write penalty for partial writes.

-B

-- 
Brandon High : bhigh at freaks.com
If violence doesn''t solve your problem, you''re not using
enough of it.

Tim Cook wrote:
> On Sun, Nov 15, 2009 at 2:57 AM, besson3c <joe at netmusician.org
> <mailto:joe at netmusician.org>> wrote:
>
>     Anybody?
>
>     I would truly appreciate some general, if not definite insight as
>     to what one can expect in terms of I/O performance after adding
>     new disks to ZFS pools.
>
>
>  
> I''m guessing you didn''t get a response because the first
result on
> google should have the answer you''re looking for.  In any case, if
> memory serves correctly, Jeff''s blog should have all the info you
need:
> http://blogs.sun.com/bonwick/entry/raid_z
>
>
Hey Tim,

I have seen and read that page several times before and after writing this.

I''ve see several comparisons to existing RAID solutions, but
I''m not
finding whether the more disks you add, the more I/O you can get, unless
I''m missing something?

> --Tim

-- 
Joe Auty
NetMusician: web publishing software for musicians
http://www.netmusician.org
joe at netmusician.org
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://mail.opensolaris.org/pipermail/zfs-discuss/attachments/20091115/24976248/attachment.html>

On Sun, 15 Nov 2009, Joe Auty wrote:
> 
> I''ve see several comparisons to existing RAID solutions, but
I''m not finding whether the more
> disks you add, the more I/O you can get, unless I''m missing
something?
Perhaps that is because "it depends" and you may or may not get
"more
I/O", depending on what "more I/O" means to you.

In some cases the I/O rates depend on decision-making in the zfs code. 
For example, while huge sequential I/O rates may be available, the zfs 
code might decide to ramp-up the prefetch rate over time so that you 
only see the huge rate for very large files.  The algorithms and 
decisions in zfs tend to change over time based on bug reports and the 
zfs implementor''s accumlated experience.

Bob
--
Bob Friesenhahn
bfriesen at simple.dallas.tx.us, http://www.simplesystems.org/users/bfriesen/
GraphicsMagick Maintainer,    http://www.GraphicsMagick.org/