Thread: I have a fusion IO drive available for testing

XFS

- Luke

We have a box outfitted with two of them we are testing.   We are testing with VxFS and ext3, but not quite ready to publish.   I will reply when we are done.


-----Original Message-----
From: pgsql-performance-owner@postgresql.org on behalf of Dave Cramer
Sent: Thu 3/26/2009 5:47 AM
To: pgsql-performance@postgresql.org
Subject: [PERFORM] I have a fusion IO drive available for testing

So far using dd I am seeing around 264MB/s on ext3, 335MB/s on ext2 write
speed. So the question becomes what is the best filesystem for this drive?

Anyone want me to run anything on it ?

Dave

On Mar 26, 2009, at 8:47 AM, Dave Cramer wrote:

> So far using dd I am seeing around 264MB/s on ext3, 335MB/s on ext2
> write speed. So the question becomes what is the best filesystem for
> this drive?
>
> Anyone want me to run anything on it ?
>
> Dave


I'd be more interested in the random io numbers.

You can do some tests with pgiosim (avail on pgfoundry) to sort-of
simulate an index scan. It just seeks and reads. It can also randomly
write and or fsync.

I'd be interested in seeing numbers for 1 proc and 10 on the fusionIO.
You have to make some file(s) for it to use first (I usually use dd to
do that, and make sure it is at least 2xRAM in size)



--
Jeff Trout <jeff@jefftrout.com>
http://www.stuarthamm.net/
http://www.dellsmartexitin.com/

On Thu, 26 Mar 2009, Dave Cramer wrote:

> So far using dd I am seeing around 264MB/s on ext3, 335MB/s on ext2 write
> speed. So the question becomes what is the best filesystem for this drive?

until the current mess with ext3 and fsync gets resolved, i would say it
would probably be a bad choice. I consider ext4 too new, so I would say
XFS or ext2 (depending on if you need the journal or not)

for the WAL you definantly don't need the journal, for the data I'm not
sure. I believe that postgres does appropriate fsync calls so is safe on a
non-journaling filesystem. the fusionIO devices are small enough that a
fsync on them does not take that long, so it may not be worth the overhead
of the journaling.

David Lang

> Anyone want me to run anything on it ?
>
> Dave
>

On Mar 27, 2009, at 1:30 PM, david@lang.hm wrote:

>
> for the WAL you definantly don't need the journal, for the data I'm
> not sure. I believe that postgres does appropriate fsync calls so is
> safe on a non-journaling filesystem. the fusionIO devices are small
> enough that a fsync on them does not take that long, so it may not
> be worth the overhead of the journaling.
>

The win for the journal on the heap is simply so you don't need to
spend $longtime fsck'ing if you crash, etc.

--
Jeff Trout <jeff@jefftrout.com>
http://www.stuarthamm.net/
http://www.dellsmartexitin.com/

On Fri, Mar 27, 2009 at 10:30 AM,  <david@lang.hm> wrote:
> On Thu, 26 Mar 2009, Dave Cramer wrote:
>> So far using dd I am seeing around 264MB/s on ext3, 335MB/s on ext2 write
>> speed. So the question becomes what is the best filesystem for this drive?
>
> until the current mess with ext3 and fsync gets resolved, i would say it
> would probably be a bad choice. I consider ext4 too new, so I would say XFS
> or ext2 (depending on if you need the journal or not)

If you're worried about the performance implications of ext3 in
data=ordered mode, the best thing to do is to mount the filesystem in
data=writeback mode instead.

If you're only using the filesystem for PostgreSQL data or logs, your
data will be just as safe except now that data and metadata won't be
forced to disk in the order it was written.

And you still get the benefit of a journal so fsck's after a crash will be fast.

XFS probably is a decent choice, but I don't have much experience with
it except on a desktop system where I can tell you that having write
barriers on absolutely kills performance of anything that does a lot
of filesystem metadata updates.  Again, not a big concern if the
filesystem is only being used for PostgreSQL data or logs.

-Dave

Dave Cramer wrote:
> So I tried writing directly to the device, gets around 250MB/s, reads at
> around 500MB/s
>
> The client is using redhat so xfs is not an option.

I'm using Red Hat and XFS, and have been for years. Why is XFS not an option with Red Hat?

Craig

craig_james@emolecules.com (Craig James) writes:
> Dave Cramer wrote:
>> So I tried writing directly to the device, gets around 250MB/s,
>> reads at around 500MB/s
>>
>> The client is using redhat so xfs is not an option.
>
> I'm using Red Hat and XFS, and have been for years. Why is XFS not an option with Red Hat?

If you report practically any kind of problem, and you're using XFS,
or JFS, or such, their "support offering" is to tell you to use a
supported filesystem.
--
let name="cbbrowne" and tld="cbbrowne.com" in name ^ "@" ^ tld;;
http://linuxdatabases.info/info/linuxxian.html
"The only thing  better than TV with the  sound off is  Radio with the
sound off." -- Dave Moon