Postgres on filer via NFS

Sorry for the newb question around filers; is there any truth in this
(old) recommendation re: NFS still?

http://www.postgresql.org/files/documentation/books/aw_pgsql/hw_performance/node11.html

If not, perhaps it should have a caveat or be removed as it turns up in
searches and threads?

Brian, looking at NetApp FAS250 for ~40GB Postgres storage

On Tue, Mar 24, 2009 at 04:36:30PM -0700, Brian Ghidinelli wrote:
>
> Sorry for the newb question around filers; is there any truth in
> this (old) recommendation re: NFS still?

Absolutely. Nothing about network-attached storage has changed in
general, and the article leaves out the fact that most people's
network pipes far under-perform direct-attached storage in terms of
the number of bits it's possible to push through the pipe in a second.

> http://www.postgresql.org/files/documentation/books/aw_pgsql/hw_performance/node11.html
>
> If not, perhaps it should have a caveat or be removed as it turns up in
> searches and threads?
>
> Brian, looking at NetApp FAS250 for ~40GB Postgres storage

DON'T! Just get a flock of small-radius SAS drives and arrange them
as a RAID-10 :)

Cheers,
David.
--
David Fetter <david(at)fetter(dot)org> http://fetter.org/
Phone: +1 415 235 3778 AIM: dfetter666 Yahoo!: dfetter
Skype: davidfetter XMPP: david(dot)fetter(at)gmail(dot)com

Remember to vote!
Consider donating to Postgres: http://www.postgresql.org/about/donate

On Tue, Mar 24, 2009 at 04:36:30PM -0700, Brian Ghidinelli wrote:
> Sorry for the newb question around filers; is there any truth in this
> (old) recommendation re: NFS still?
>
> http://www.postgresql.org/files/documentation/books/aw_pgsql/hw_performance/node11.html
>
> If not, perhaps it should have a caveat or be removed as it turns up in
> searches and threads?
>
>
> Brian, looking at NetApp FAS250 for ~40GB Postgres storage

Depending on your IO needs, running via NFS on a Netapp is perfectly feasible
and has some distinct advantages (fast backups using snapshots are the most
compelling):

http://archives.postgresql.org/sfpug/2008-06/msg00007.php

make sure you have enough NFS client threads...

Adi

David Fetter wrote:
> DON'T! Just get a flock of small-radius SAS drives and arrange them
> as a RAID-10 :)

Thanks for the feedback. Jeff Frost once painfully proved to me how
fast the new drives and DAS can be. There are obviously some advantages
to a filer (NetApp in this case) like the cloning, management and
sharing it among boxes.

> Absolutely. Nothing about network-attached storage has changed in
> general, and the article leaves out the fact that most people's

So basically you need a sane NFS config and it's OK then? We can do FCP
on this NetApp so speed I don't think will be an issue. Plus our
database will fit into RAM so management/availability "features" is as
important as speed. I know Hi5 is all NetApp so apparently it *does*
work. I just wondered if they had magic genies. :)

> http://archives.postgresql.org/sfpug/2008-06/msg00007.php

Adi, I saw this thread previously; the diverging opinions out there are
what prompted my question. Thanks for the input!

Brian

Brian,

See blog discussion. NFS is just like iSCSI, only worse.
http://it.toolbox.com/blogs/database-soup/the-problem-with-iscsi-30602

If you *do* to this because management is more important than
performance, then budget for $2k or so worth of tuning & troubleshooting
services, or make sure that that's included by NetApp. You'll also need
NetApp's proprietary NFS tech to make it work at all.

Also note that lag time isn't such a problem for data warehousing as it
is for OLTP. And the bandwidth issues are solvable by using 10gE, but
that significantly increases the cost of your solution.

--Josh