zfs discuss - Dec 2007 - LUN configuration for disk-based backups

Hello,

We have a StorageTek FLX280 (very similar to a 6140) with 16 750 GB SATA drives
that we would like to use for disk-based backups. I am trying to make an
(educated) guess at what the best configuration for the LUN''s on the
FLX280 might be.

I''ve read, or at least skimmed, most of the "ZFS Best Practices
Guide" over at solarisinternals.com, which has some great information;
however, I still do not feel like I have a good understanding of the interaction
between ZFS and a disk array.

More specifically, I am concerned about the number of IOP''s that each
drive and/or LUN will be able to handle. Seagate lists an average seek time of
9ms, and an average rotational latency of 4.16ms for these drives. By my math,
each drive should be capable of 76 IOP''s in a worst case scenario; i.e.
completely random I/O.

These drives support native command queuing, and the controllers on the FLX280
have a battery-backed cache, so I would _assume_ that they are also capable of
reordering I/O op''s to improve throughput to the disks.

So, the question is whether or not worst-case IOP''s are even relevant.
If my assumptions about the controllers on the FLX280 are correct
(documentation?), then it seems like we could use RAID-Z to get the throughput
we''re looking for. If not, we may have to go with RAID-1.

Anyone have any thoughts on this?
 
 
This message posted from opensolaris.org

On Fri, 14 Dec 2007, Andrew Chace wrote:

[.... reformatted .... ]
> Hello,
>
> We have a StorageTek FLX280 (very similar to a 6140) with 16 750 GB 
> SATA drives that we would like to use for disk-based backups. I am 
> trying to make an (educated) guess at what the best configuration 
> for the LUN''s on the FLX280 might be.
>
> I''ve read, or at least skimmed, most of the "ZFS Best
Practices
> Guide" over at solarisinternals.com, which has some great 
> information; however, I still do not feel like I have a good 
> understanding of the interaction between ZFS and a disk array.
>
> More specifically, I am concerned about the number of IOP''s that 
> each drive and/or LUN will be able to handle. Seagate lists an 
> average seek time of 9ms, and an average rotational latency of 
> 4.16ms for these drives. By my math, each drive should be capable of 
> 76 IOP''s in a worst case scenario; i.e. completely random I/O.
>
> These drives support native command queuing, and the controllers on 
> the FLX280 have a battery-backed cache, so I would _assume_ that 
> they are also capable of reordering I/O op''s to improve throughput
> to the disks.
>
> So, the question is whether or not worst-case IOP''s are even 
> relevant. If my assumptions about the controllers on the FLX280 are 
> correct (documentation?), then it seems like we could use RAID-Z to 
> get the throughput we''re looking for. If not, we may have to go
with
> RAID-1.
>
> Anyone have any thoughts on this?
Since ZFS makes it so quick/easy to create storage pools and 
filesystems, the simplist way to determine your optimum config is to 
conduct a set of experiments - using your data and your applications.

Bear in mind that no one ZFS config is ideal for every user data 
application scenario - you may wish to consider 2 or more storage 
pools with different configurations that will be a best fit your 
requirements - given that you may have several different data sets 
with different characteristics.  Then there is ZFS compression - which 
might really help if your data is highly compressible.

Also ensure that you have sufficient network bandwidth into the ZFS 
backup server.

Regards,

Al Hopper  Logical Approach Inc, Plano, TX.  al at logical-approach.com
            Voice: 972.379.2133 Fax: 972.379.2134  Timezone: US CDT
OpenSolaris Governing Board (OGB) Member - Apr 2005 to Mar 2007
http://www.opensolaris.org/os/community/ogb/ogb_2005-2007/
Graduate from "sugar-coating school"?  Sorry - I never attended! :)