Juan Casero wrote:
> I have read the caveats posted on the SGI OSS web site regarding
> XFS but I still want to get some opinions from the members on the
> list (especially the developers) regarding the readiness of XFS
> for some serious work ... <cut> ... Currently all the systems use
> ext2 file systems. My questions is simple; how long before the XFS
> developers feel it might be safe to put XFS to in an environment
> like this? Certainly the journaling features of XFS is the most
> important thing we need but there is also the performace issue to
> consider (I believe XFS is slightly faster than ext2) and the
> maturity of the code.
As a non-developer/sysadmin with about 9 months of Ext3 usage (6
months on production fileservers), and a recent adopter of XFS
(testing right now), I can offer the following pointers and advise:
1. Personal testing and experience is the only way to find out if
something is _right_for_you_. Again, I tested Ext3 for 3 months on
less important servers and workstations before adopting it on my
heavily used workstations and main file servers.
2. Your JFS options will be limited by your system application and
kernel (even GLibC). One JFS may work great for some applications
(e.g., ReiserFS for Linux workstations), but be a total no-no on
others (e.g., ReiserFS for Linux NFS fileservers). And some JFS are
kernel/GLibC-dependent (Ext3 is 2.2-only right now, and kernel 2.4 +
GLibC 2.2 -- i.e. RedHat 7 -- is recommended for XFS).
3. If you want my "'biased', 'quickie' breakdown" of the 4 major
JFS for Linux -- they are as follows:
Tweedie@RedHat's Ext3 is remedially "evolutionary"
(fully reversable, it's Ext2 with journaling slapped on)
Namesys' ReiserFS is very "revolutionary" for Linux
(and hasn't been proven anywhere but on Linux, if that)
SGI's XFS is a port to Linux
(powerful, feature-rich yet still legacy UNIX compatible)
IBM's JFS seems to be no sparsely available?
(Possible IBM hardware/support-focused?)
4. Performance is a non-issue for web servers IMHO
Unless you have a fast connection, or a serious database back-end,
your bottleneck is not going to be the filesystem, but the Internet
connection. I could be wrong, but I wouldn't worry about
performance.
With that said, XFS is the fastest of the four JFS in most apps, and
Ext3 is just as fast as Ext2 for _reads_. So its depends on how
much disk I/O and static v. dynamic content you will have (or be
generating in the case of the latter).
5. Personal Experiences
- NFS (and SMB) Fileservers with Linux 2.2 + Ext3 + NFS3
I have production NFS fileservers, including Linux (which are also
SMB fileservers via Samba). As such ReiserFS is NOT an option (not
even with the 2.4.x patches for kNFSd IMHO). For kernel 2.2, I
ended up adopting Ext3 (along with the NFS3 backport to 2.2) for a
good, solid and reliable JFS. I _always_ run in the V1 mode (full
data journaling) instead of trying to use the newer meta-data-only
modes. That also means write performance is effectively _halfed_
with Ext3 (for reads, Ext3 = Ext2). [ It's definately NOT the
performance route for fileservers! ]
- Kernel 2.4 limits your options, but introduces XFS
Now when it comes to kernel 2.4, Ext3 is finally getting ported by
Tweedie (the man is bogged down with a crapload from my
understanding -- and waited for the official 2.4.0 release before
starting to address it, which is probably smart). That leaves
ReiserFS and XFS. Unlike XFS, ReiserFS is also supported on kernel
2.2 (albeit somewhat limited in features though).
- ReiserFS _may_ work for you
Now you may be able to get away with ReiserFS. It is designed to be
dynamically self-upgrading (including from Ext2) which may be a
nice, gradual way for you to upgrade over time (making lesser
important partitions ReiserFS, and converting others to Ext2
later). Most of ReiserFS' issues (other than the fact that its a
from-the-ground-up filesystem and relatively new compared to
Ext2/Ext3 and XFS) is with NFS services. If you aren't going to use
kNFSd on your web servers (for sharing information between each
other), then it might be a viable option.
- XFS is the 2.4 "bomb" and its most feature rich/working on Linux
On the other hand, XFS is here for kernel 2.4 and has the richest
feature set, the most capabilities working _now_ in 2.4, and I have
a bunch of XFS-enable kernel 2.4.2 and support RPMs "ready to plop
down" on a stock RedHat 7.0 install (URL below). As such, it might
be _easiest_ to just setup a couple of test RedHat 7.0 + my XFS RPMs
and run for a couple of months (FYI, I plan to release new XFS
"snapshot" RPM releases from CVS every week or two) serving out your
least critical stuff.
- But only personal, first-hand testing will tell you
Again, you need to test it yourself, in your environment, to see if
it works for you. After 3 months, you'll know if it works for you.
-- TheBS
--
Bryan "TheBS" Smith, Engineer CONTACT INFO
***********************************************************
Chat: thebs413 @ AOL/MSN/Yahoo (see http://Everybuddy.com)
Email: mailto:thebs@xxxxxxxxxxxxxxxxx,thebs@xxxxxxxxxxx
|