Filesystem benchmarks

[Tom 'Needs A Hat' Mitchell]

On Thu, May 13, 2004 at 07:46:32AM +0800, Christopher Chan wrote:
> Date: Thu, 13 May 2004 07:46:32 +0800
> From: Christopher Chan <cchan at outblaze.com>
> To: Robin.Laing at drdc-rddc.gc.ca,
>         For users of Fedora Core releases <fedora-list at redhat.com>
> Cc: 
> Subject: Re: Filesystem benchmarks
> Reply-To: For users of Fedora Core releases <fedora-list at redhat.com>
> 
> Robin Laing wrote:
> >I don't know how many people have seen this but it may be of interest or 
....
> >Benchmarking Filesystems
....
> >This article covers ext2, ext3, JFS, REISERFS and XFS.
....

> A far more interesting one can be found at 
> http://untroubled.org/benchmarking/2004-04/
> 
> which simulates real world conditions for mta boxes with recent kernels 
> and not with a kernel that is not going to receive much attention anymore.

yes it does look more interesting:

    "Benchmark Goals

    "The goal of this set of benchmarks is to determine which of the
    leading Linux filesystems (ext2, ext3, ReiserFS, and XFS) offer the
    best performance when used for accepting maildir deliveries. The
    resulting system should be a good balance of delivery and retrieval
    performance under potentially high concurrent filesystem load."

This is interesting because it addresses a real and specific
use of a filesystem.   If your usage pattern is different
then a different data set needs to be generated.

It is a good example of measure the need and shape the benchmark to
match.  Most vendor driven benchmarks focus on what they do well not
on what you need them to do well ;-)


> Add to that that things have changed so much in 2.6 from 2.4 that fs 
> code is substantially different. eg: XFS in 2.4 is different from XFS in 2.6

Yes...

And with a quick look the benchmark seems to make some simplifying
assumptions that may or may not match your application needs.  He also
did not report or tune any of the file systems mkfs params.

For example in the mke2fs man page: If omitted, mke2fs block- size is
hueristically determined by the file system size and the expected
usage of the filesystem (see the -T option)."  So we do not know
exactly how his ext3 filesystems (and others) were tuned.  The
hueristics could be researched and we could discover the resulting
parameters but at this moment we do not exactly know.

To his credit he clearly exposes his method, data and goals.
Very good stuff!


-- 
	T o m  M i t c h e l l 
	/dev/null the ultimate in secure storage.