zfs discuss - May 2009 - Any fragmentation issue with zfs, espeically loaded with lots of snapshots

Hi,



I''d like to know more about this topic.  However, googling
doesn''t give
me much.  If you guys know of any info, please point me to it.



I understand zfs has high performance write, even with lots snapshots.
How about read performance?  To me, it will be a very fragmented due to
the COW.  And no matter how careful you allocate the blocks, you can
only put a SHARED block in the proximity of one particular snapshot and
to other snapshots, the shared block won''t be in proximity - a
fragmentation situation.



Appreciate any help,

James



_____________________________________________________________________________

ATTENTION:

The information contained in this message (including any files transmitted 
with this message) may contain proprietary, trade secret or other 
confidential and/or legally privileged information. Any pricing 
information contained in this message or in any files transmitted with 
this message is always confidential and cannot be shared with any third 
parties without prior written approval from Syncsort. This message is 
intended to be read only by the individual or entity to whom it is 
addressed or by their designee. If the reader of this message is not the 
intended recipient, you are on notice that any use, disclosure, copying or 
distribution of this message, in any form, is strictly prohibited. If you 
have received this message in error, please immediately notify the sender 
and/or Syncsort and destroy all copies of this message in your possession, 
custody or control.
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://mail.opensolaris.org/pipermail/zfs-discuss/attachments/20090520/60756f2b/attachment.html>

Wong, James (Nagi Long) wrote:
>
> Hi,
>
>  
>
> I?d like to know more about this topic.  However, googling doesn?t 
> give me much.  If you guys know of any info, please point me to it.
>
>  
>
> I understand zfs has high performance write, even with lots snapshots. 
>  How about read performance?  To me, it will be a very fragmented due 
> to the COW.  And no matter how careful you allocate the blocks, you 
> can only put a SHARED block in the proximity of one particular 
> snapshot and to other snapshots, the shared block won?t be in 
> proximity ? a fragmentation situation.
>
To begin, you should look at how blocks are allocated.  Jeff Bonwick
wrote a good blog on this:
http://blogs.sun.com/bonwick/entry/zfs_block_allocation

Next, you''ll need to understand that high performance and magnetic
hard disks are becoming mutually exclusive.  High performance today
is owned by SSDs, which have no seek/rotate latency and tend to be
able to saturate the channel bandwidth.  The "fragmentation" problem
with HDDs is very different than "fragmentation" in SDDs. To
understand how that is leveraged for the best price/performance mix,
you''ll want to google for "hybrid storage pools" where there
are many
good reference.

The next interesting area is where the allocation policy could go in
the future -- I''m not convinced that the HDD allocation policy in
place today is the best for SSDs. It certainly works for SSDs, but
there may be some improvements possible.  So much research, so
little time...

<shameless plug>
I''ll be getting into some of this sort of thing in some detail at the
USENIX Technical Conference 2009 ZFS Tutorial.  You''re
welcome to attend, and bring some friends :-)
http://www.usenix.org/events/usenix09/training/
</shameless plug>
 -- richard

On Wed, 20 May 2009, Wong, James (Nagi Long) wrote:
>
> I understand zfs has high performance write, even with lots snapshots.
> How about read performance?  To me, it will be a very fragmented due to
> the COW.  And no matter how careful you allocate the blocks, you can
> only put a SHARED block in the proximity of one particular snapshot and
> to other snapshots, the shared block won''t be in proximity - a
> fragmentation situation.
This is indeed an interesting issue.  In practice, ZFS writes 
essentially unfragmented files for normal sequential writes, which 
results in an unfragmented read.  It uses a "slab" allocator which 
pre-allocates large runs of blocks at a time.  The default ZFS block 
size is quite large (128K) so worst-case fragmentation causes much 
less disk seeking than filesystems using tiny (4K or 8K) blocks.  ZFS 
buffers data to be written for considerable time (up to 30 seconds) in 
the hope that the data can be written more contiguously when it comes 
time to write it to disk.

Fragmentation becomes an issue for applications which update 
individual ZFS blocks (like a database) rather than re-writing the 
file.  Fragmentation also becomes an issue when the pool is extremely 
full.  Fragmentation is an issue when the block size is set very small 
(e.g. 8k) but usually this small block size is used intentionally for 
database purposes.

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