Re: Filesystem Preference?

Roy Sharp wrote:
> I'm getting ready to make the switch from Fedora to Gentoo on my
> laptop. I was wondering which file system you all prefer and why
> (ext3, reiserFS, or XFS).
> Another question. Is it possible to compile the driver w/ 16k stack.
> My wireless card requires it (Broadcom w/ ndiswrapper). Under Fedora
> I downloaded the 16k kernel rpm,  will  this  be possible to configure
> when compiling.
> 
> Thanks for your help, and wish me luck.
> 
> Roy
> 

OMG - FS Questions are nearly as religious as Distro debates.

Hands down, all useless opinions aside, the best choice is clearly XFS.

But, I don't know what you need.

I use XFS for everything.   I have tried ext[23], reiser3, and jfs.

ext2 is the most stable and tested FS.   Not journaled, so long rebuilds 
after power failure / hardware / stupid user, etc.

ext3 has a 32k/directory entry problem.  You cannot go over that,  I 
don't like it.   I have tried it a bunch and never been satisfied w/ how 
it felt.   It has had all the problems common to a 'new journaled fs'. 
  Beyond that, it has had an alarming series of "everything was ok, but 
then I ran fsck.ext3 and all this stuff disappeared" (aka silet 
corruption).   I don't know about recent [0 - 6 months] as far as what 
it does now.  It also seems to prefer running fsck after crashes - so 
not real good.   On the plus side, it is the only FS that offers a 'full 
data journal' mode.  All others only journal the fs meta-data.  It is 
debatable whethey full data journaling is important or not.  [Default 
mode is w/o data journal]

reiser3 is a great research project.  It has stabilized.   I don't much 
care for it.  I did not have good luck at all.  I lost some data, lost a 
lot of time, and had some generally really bad wierd problems.   That 
was in the early days of 2.4, so who knows.  It requires special options 
on yopur boot drive.  In theory it makes everything more space 
efficient.  I don't really know that you can quantify the space savings 
though.   Hans and the guys talked it up, but I don't know that it has 
any pay off.

jfs, I don't really know that much about.  I ran it long enough to post 
about a JFS bug where the developers said "Known issue, good luck".  No 
one really cared.  One client was preparing to migrate to it, then thier 
fs went totally b0rked and they lost all of the stuff on the test 
partition.   I don't know if it was really JFSs fault, but I lost the 
taste for it due to the developers not caring.   In all the benchmarks, 
JFS and XFS are neck and neck.   JFS wins for being less CPU intensive.

xfs, the only one I know of that actually has a defragmenter.  [It 
doesn't need it anymore than the rest of them do, but it still has it.] 
    I have partitions on some laptops as small as 2 gig.   There are 
people on the XFS list who consider 1.5TB only a 'jumping off point' to 
real "large" storage.  SGI actually commercially supports XFS.   They 
have some sort of commercial version, which I believe is actually for 
clustering.  This isn't an ad, and I don't know much about it.  I have a 
900GB array that I use with it.

I monitor the XFS list, so I hear about the stories of problems.  One 
guy ran into corruption after making 20 directories, then spawning 20 
threads to do various operations, then after about 164 cycles he powered 
his machine off.  He said he did this 200 times, and he seemed suprised 
about some corruption.

In the end, he successfully repaired and the moral of the story was that 
disks have a small write cache, and that reduces the survivability of 
_any_ journaled filesystem, especially if you repeatedly post off the 
box under high load.

ext3 [in default mode], reiser, jfs and xfs all journal fs meta-data. 
The goal is to primarily preserve the FS, and second guarantee that 
writes are either 100% successful, or 100% failed.  You can get some 
occasional corruption from power-cycles, and the like under high load, 
but nothing too bad.

Journals and the FS structures are vulnerable to write caching of all 
kinds.   They need to _know_ that write X is happening before write Y. 
There are problems if you have out-or-order writes to disk.

Write barriers are now in 2.6.8, and I know that ext3 is already using 
them.  I don't know about any of the others.


===


So, after all that.  I feel safe depending on XFS.  It can handle HUGE 
files, and it dsoes fine at handling small ones.  It has good support, 
an active IRC channels, and I have 30+ boxes running it.

js






_______________________________________________
PDXLUG mailing list
[email protected]
http://pdxlug.org/mailman/listinfo/pdxlug

_______________________________________________
PDXLUG mailing list
[email protected]
http://pdxlug.org/mailman/listinfo/pdxlug

-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.2.4 (GNU/Linux)
Comment: Using GnuPG with Thunderbird - http://enigmail.mozdev.org

iD8DBQFBQm5UcweM6t71VHsRAosVAJ9c2a9xtRREESAX2Xthj27N5zzBFwCg0YAV
e3jt7aXLwppKypxaP2jxhyE=
=XLrO
-----END PGP SIGNATURE-----