Re: Opinions on ext3 vs XFS vs reiserfs for LAMP server
On 8/23/07, Neil Gunton <neil@nilspace.com> wrote:
> I am wondering if anyone has any real-world advice on the best
> filesystem to use for an AMD64 LAMP server. I know that the different
> systems have their pros and cons, but a lot of the comparisons out there
> seem to be rather old. So I'll describe below exactly what it is I'm doing.
I don't see that reality has changed terribly much over the last few
years, except in ways that make choices other than ext3 look worse
than they used to.
The last time I did a write-heavy benchmark, I found that JFS was, by
a small margin, fastest, with XFS a bit slower, and ext3 in behind.
<http://archives.postgresql.org/pgsql-general/2003-11/msg01692.php>
I did no evaluation of ReiserFS, at the time, because there were known
issues with regard to reliability, and they had some evident
anti-features for DBMS work.
I see no indication that things have changed in the last four years in
material ways vis-a-vis performance. And performance is only a
primary issue if you're an idiot; it should take a back seat to
reliability. THAT is then where things get interesting:
1. In 2003, you couldn't get support from RHAT on RHAS/RHES if you
were using any filesystem other than ext3. It is not evident that
this has changed; I see no reason for them to have added JFS or XFS to
the list, and I'm quite sure ReiserFS hasn't been added :-).
If we told them we had a problem with something, they'd ask for system
details, notice that we had a JFS filesystem mounted, and then tell us
that they declined to offer any more support until we got rid of the
non-vendor-provided kernel components.
2. SuSE was a bit more open to supporting other filesystems; they
would not actually tell us to "go away" about JFS or ReiserFS.
Neither 1. nor 2. are directly relevant to deployment of Debian/AMD64,
but what risks vendors are willing to take should be considered.
3. I have seen filesystems lost to corruption on all of [JFS, XFS,
and ReiserFS], so I have at least vague, anecdotal evidence against
their use.
4. If you look at ongoing development efforts, you'll find that:
a) IBM isn't working all that actively on JFS; they have more staff
devoted to ext3 (e.g. - such as Ted T'so)
b) Is there anything left of SGI to work on XFS?
c) Do I have to say anything about recent happenings surrounding Hans Reiser?
I don't see the heavy usage and decent development patterns there to
be causing any of these alternative filesystems to become compellingly
more reliable in their integration with the Linux kernel.
Mind you, by your other decisions, you may have pre-selected some set
of risks. It may be that the risks associated with selecting
(XFS|ReiserFS|JFS) are not material in comparison with (say) the risks
that you have already accepted that are associated with the paucity of
data integrity controls within MySQL(tm).
If you already have a sizable "whack of risk," filesystem selection
may not add materially to the pre-existing risks.
Reply to: