On 6/16/06, Hemlock <email@example.com> wrote:
Seeking some advice on which filesystem to use.
I've been primarlity an ext3 user, because thats all I've really
ever known of, but would consider trying something else.
Plus, never used anything other than 32bit Debian since slink.
I've read some articles googling for xfs, ext3 and jfs and such.
Leaning towards xfs maybe?
This will be for a AMD64 sarge box used as a file, web, email and proxy
server. Also running software raid.
Thanks for your advice.
At work (Afilias; we operate registries for various TLDs including
.org, .info, and such), we did a bunch of work a couple years ago in
order to try to figure out what filesystem to prefer for our Linux
The "short list" was XFS, JFS, and ext3.
Performance-wise, I found that (for a heavy DBMS load) JFS was
*slightly* preferable to XFS which was a fair bit quicker than ext3.
Unfortunately, we then ran into a couple problems...
1. We wanted to have commercial distribution support. RHAT wouldn't
provide meaningful tech support if you're not running ext3; as soon as
anything else was mentioned, they'd emit the mantra "use a supported
SuSE wouldn't pester us when we used JFS, but warned us that there
were known corruption issues (unfixed) with XFS on Opteron.
2. Heavy DBMS load on large (>2TB) filesystems has been able to
corrupt pretty much any of the fses.
Important systems are now on AIX/JFS2, as a result, even though AIX is
its own barrel of dung-flinging monkeys.
It's not completely evident where the problems lie; it's somewhere
between the assorted layers of:
- FibreChannel HBA drivers talking to
- SCSI APIs talking to
- VFS talking to
There's something deeply wrong there; with the current rush to "Git"
code thrown at the various Linux kernel lieutenants, I am not
optimistic that there will be fixes found to any of this soon.
3. I've been watching where efforts + changes have been going, and
ext3 seems to be the only thing seriously supported...
JFS is infrequently updated, and it's most interesting that the "heavy
hitters" in ext3 development (Ted T'so, notably) are at IBM.
XFS is an SGI project, and the recent bankruptcy proceedings (chapter
11 reorg) do not fill me with confidence.
Note that at OLS, last summer, there were substantial talks about
ext3; there were *none* on XFS or JFS.
That points me at ext3 being the thing of primary ongoing interest...