Thread: Linux filesystem shootout

Linux filesystem shootout

From
Shridhar Daithankar
Date:

Re: Linux filesystem shootout

From
Grega Bremec
Date:
...and on Thu, Oct 09, 2003 at 04:42:53PM +0530, Shridhar Daithankar used the keyboard:
>
> http://www.ussg.iu.edu/hypermail/linux/kernel/0310.1/0208.html
>
> Shridhar

My $0.1:

I just stumbled across an interesting filesystem comparison table today,
comparing ext2/ext3/reiser/reiser4/jfs/xfs on a single UP P2/450 machine
with an old UDMA2 Seagate.

Now however archaic this box may have been, I think that the tests still
bear some objectivity, as it's a comparison test and not some "how much
can we squeeze out of xyz" type of bragging.

The tests were done using bonnie++ and IOZone and are essentially just a
couple of tables listing the average results achieved by each of those
tests.

Also, ext3, reiser and reiser4 were tested in a couple of different
configurations (reiser4 extents, reiser notail, ext3 journal, ordered and
writeback mode).

Oh, i shouldn't forget - the address is http://fsbench.netnation.com/ :)

Cheers,
--
    Grega Bremec
    Sistemska administracija in podpora
    grega.bremec-at-noviforum.si
    http://najdi.si/
    http://www.noviforum.si/

Attachment

Re: Linux filesystem shootout

From
Grega Bremec
Date:
I should at least read the URLs before re-posting info.

My bad, I'm utterly sorry about this... :-(

Cheers,
--
    Grega Bremec
    Sistemska administracija in podpora
    grega.bremec-at-noviforum.si
    http://najdi.si/
    http://www.noviforum.si/

Attachment

Re: Linux filesystem shootout

From
Kaarel
Date:
<br /><blockquote cite="mid20031009121124.GA7038@elbereth.noviforum.si" type="cite"><blockquote type="cite"><pre
wrap=""><aclass="moz-txt-link-freetext"
href="http://www.ussg.iu.edu/hypermail/linux/kernel/0310.1/0208.html">http://www.ussg.iu.edu/hypermail/linux/kernel/0310.1/0208.html</a>

Shridhar   </pre></blockquote><pre wrap=""></pre></blockquote> I feel incompetent when it comes to file systems. Yet
everybodywould like to have the best file system if given the choice...so do I :) Here I am looking at those tables
seeingJFS having more green cells than others. The more green the better right? So based on these tests JFS ought to be
theone?<br /><br /> Kaarel<br /> 

Re: Linux filesystem shootout

From
Shridhar Daithankar
Date:
Kaarel wrote:
>>>http://www.ussg.iu.edu/hypermail/linux/kernel/0310.1/0208.html
>>>
>>>Shridhar
>>>
>>>
> I feel incompetent when it comes to file systems. Yet everybody would like to
> have the best file system if given the choice...so do I :) Here I am looking at
> those tables seeing JFS having more green cells than others. The more green the
> better right? So based on these tests JFS ought to be the one?

Yes and no. Yes for the results. No for the tests that weren't run.

Database load is quite different. Its mixture of read and write load with a
dynamics varying from one extreme to other, between these two.

All it says that if you want to choose a good file system for postgresql, look
at JFS first..:-)

  Besides all the tests were done on files file bigger than 1GB. If single file
size is restricted to 1GB, it might produce a different result set. And
postgresql does not exceed 1GB limit per file.

So still, quite a few unknowns there..

Best thing could be repeat those benchmarks on $PGDATA with your live data
inside it. It could mimmic the load pretty well..

  Shridhar


Re: Linux filesystem shootout

From
Bill Moran
Date:
Kaarel wrote:
>
>>>http://www.ussg.iu.edu/hypermail/linux/kernel/0310.1/0208.html
>>>
>>>Shridhar
>>>
> I feel incompetent when it comes to file systems. Yet everybody would
> like to have the best file system if given the choice...so do I :) Here
> I am looking at those tables seeing JFS having more green cells than
> others. The more green the better right? So based on these tests JFS
> ought to be the one?

Those tests seem to align with the ones I did recently:
http://www.potentialtech.com/wmoran/postgresql.php#results

There were less filesystems involved, and the data is less comprehensive,
but probably a little easier to understand (i.e. -> fastest filesystem
at the top of the graph, slowest at the bottom).

I've been telling people that JFS is fastest.  This is definately
oversimplified, since the "shoot out" shows that it's not _always_
fastest, but for people who just want to make a good initial choice,
and won't do their own testing to find out what's fastest in their
configuration (for whatever reason), I think JFS is the safest bet.
Since it's a journalling filesystem as well, it should have good
recoverability in the even of catastrophy, but I haven't tested
that.

--
Bill Moran
Potential Technologies
http://www.potentialtech.com


Re: Linux filesystem shootout

From
"scott.marlowe"
Date:
On Thu, 9 Oct 2003, Shridhar Daithankar wrote:

> Kaarel wrote:
> >>>http://www.ussg.iu.edu/hypermail/linux/kernel/0310.1/0208.html
> >>>
> >>>Shridhar
> >>>
> >>>
> > I feel incompetent when it comes to file systems. Yet everybody would like to
> > have the best file system if given the choice...so do I :) Here I am looking at
> > those tables seeing JFS having more green cells than others. The more green the
> > better right? So based on these tests JFS ought to be the one?
>
> Yes and no. Yes for the results. No for the tests that weren't run.
>
> Database load is quite different. Its mixture of read and write load with a
> dynamics varying from one extreme to other, between these two.
>
> All it says that if you want to choose a good file system for postgresql, look
> at JFS first..:-)
>
>   Besides all the tests were done on files file bigger than 1GB. If single file
> size is restricted to 1GB, it might produce a different result set. And
> postgresql does not exceed 1GB limit per file.
>
> So still, quite a few unknowns there..

Absolutely.  For instance, one file system may be faster on a RAID card
with battery backed cache, while another may be faster on an IDE drive
with write cache disabled, while another may be faster on software RAID1,
while another might be faster on software RAID5.

If you haven't tested different file systems on your setup, you don't
really know which will be faster until you do.