Performance Difference on UFS and ZFS

Ivan Voras ivoras at freebsd.org
Fri Nov 30 12:19:37 UTC 2012


On 28/11/2012 16:54, Metin Döşlü wrote:
> Hi Sergey,
> 
> I tested it on a cc1.4xlarge EC2 instance, here is the specs:
> 
> 23 GiB of memory
> 33.5 EC2 Compute Units (2 x Intel Xeon X5570, quad-core “Nehalem” architecture)
> 1690 GB of instance storage
> 64-bit platform
> 
> I installed PostgreSQL from its port on FreeBSD. I didn't do any
> tuning for PostgreSQL or FreeBSD. Data access pattern consists of
> completely from sequential reads such "select count(*) from
> table_name". I measured performance with PostgreSQL's timing option.
> As as side note; all queries are served from memory, so there were no
> disk usage for these tests.

As others said - this is interesting and unexpected. Are you sure
everything is the same across benchmarks? Since you are running on a
virtualized platform, it may be that other users of the same storage
pool "steal" your IO performance.

I did a benchmark with PostgreSQL and ZFS vs UFS a couple of years ago,
and the conclusion was that, once tuned, the performance is very
similar, with ZFS being slightly better.

Since you are testing read-only sequential IO, can you run an
alternative test with some other benchmark such as bonnie++?


-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 260 bytes
Desc: OpenPGP digital signature
URL: <http://lists.freebsd.org/pipermail/freebsd-fs/attachments/20121130/bec21bfe/attachment.sig>


More information about the freebsd-fs mailing list