Re: Installation of xfs in RH 7.1 Stop this JFS holy war before
it starts!
Florin Andrei wrote:
> Ext3 is really lame. It's just Ext2 with some journalizing
> stuff plugged into. Think of it as "proof-of-concept"
> software. But nothing more. Period.
You know, I flammed Ext3 back spring of 2000 because of its
"performance issues." Then I got sick of the stupid fsck times --
stupid consultants setup my main fileserver with 1KB blocks so it
took 90 minutes to fsck. With ReiserFS and its kNFSd issues, it was
NOT an option. So I adopted VALinux's kernels with Ext3 and NFS3.
0 downtime, 0 crashes. Ext3 in full data journaling (v1 mode) is
flawless, because it is barely any different from proven Ext2. So,
from a _production_UNIX_network_ perspective, Ext3 is *NOT* just a
"proof-of-concept", it works -- and has for almost a year (actually,
rpmfind has been using it since January of 2000)!
Unfortunately it is only available for 2.2. Plus Ext3 is still Ext2
with all its limitations. So no, Ext3 is NOT "the future" of JFS'
for Linux. But, again, it worked _yesterday_ for 2.2 kernels. You
can argue features and capabilities and "who is better" until you
are blue in the face, but Ext3 has been working on my production
network for 9 months. I've even had a major, physical disk failure
which Ext3 was able to recover from -- by dropping down to a full
Ext2 fsck. Using various UNIX and NT versions and flavors over the
years, I trust Ext2's fsck more than just about any other FS,
journaling or otherwise (ever see an NTFS filesystem mis-read its
journal? -- NASTY!).
SO LET'S STOP THIS "HOLY WAR" BEFORE IT CONTINUES!
> Only XFS and ReiserFS are, at this moment, good, stable
> and fast jfs's for Linux.
For the most part, it is NOT an apple-to-apples comparison. Let me
make it as "simple" as I can:
Ext3 in v1 mode (full-data journaling). Basically Ext2, with
journaling slapped on -- in full-data journaling mode, write
performance is effectively half of Ext2. Reversable back to Ext2,
Ext3 can drop down to a full Ext2 fsck if things get "bad" (although
you'll want to upgrade e2fsprogs to 0.19+ for Ext3 "awareness").
Anything that works with Ext2 works with Ext3. Evolutionary = few
issues, but also means continued limitations. Currently a 2.2-only
solution. [ But do *NOT* use it in v2 mode, aka "meta-data
journaling" -- I've had data loss -- be sure to use "-o
data=journal" with Ext3 v0.0.4+ to force full-data journaling. ]
ReiserFS is a new FS, on any OS, period. Depending on who I've
talked to, and the limited time I have run it, ReiserFS is solid, or
deadly. Most people running ReiserFS in the stock 2.4.1+ kernels
have had good experiences, _unless_ they use kNFSd. I've heard of
issues even with the kNFSd patches added. As such, I've never run
it much. Either way, ReiserFS has a constantly changing structure
from version to version (although I heard the "branch" for the stock
2.4.x kernel is not, because Linus would not accept it until it was
branched with a fixed structure?). Plus it's use of inodes are
non-traditional (which causes issues with kNFSd and other
applications). For Samba-only servers, ReiserFS is probably fine.
XFS is a proven, true 64-bit JFS on Irix. Now ported to Linux, XFS
has the riches feature set of any major JFS project for Linux (at
least according to several articles I've read that break down the
features of the big 4). Unlike ReiserFS, XFS keeps the inode
meta-data structure in a traditional, kNFSd-compatible format.
Although this leads to slower performance in a few cases, it is
still much faster than Ext2 (and even ReiserFS at a few operations)
and the best in compatibility with services.
I have no experience with IBM's JFS, and I've been incorrect when
I've commented on it in the past (like the fact that it is GPL, and
not IPL like I previously though).
-- TheBS
--
Bryan "TheBS" Smith chat:thebs413 @AOL/MSN/Yahoo
Engineer mailto:b.j.smith@xxxxxxxx,thebs@xxxxxxxxxxx
********************************************************
"Linux will do for applications what the Internet did to
networks" -- Sam Palmisano, IBM Chief Operating Officer
|