On Tuesday 26 February 2002 2:46 pm, you wrote:

I had to compile my own 2.4.10 kernel to get XFS support under SuSE 7.3. Does the 2.4.16 SuSE kernel support XFS?

I believe not. XFS is a big, intrusive patch. I don't think SuSE are using it for production kernels at the moment. -- 4:06pm up 13 days, 7:47, 1 user, load average: 0.05, 0.06, 0.03

On Tue, Feb 26, 2002 at 07:33:12PM +0100, Gerhard den Hollander wrote:

* Derek Fountain <fountai@hursley.ibm.com> (Tue, Feb 26, 2002 at 04:07:39PM +0000)

...

On Tuesday 26 February 2002 2:46 pm, you wrote:

...

I had to compile my own 2.4.10 kernel to get XFS support under SuSE 7.3. Does the 2.4.16 SuSE kernel support XFS?

I would recommened ext3 as a journalling fs. It might not be as fast as reiserfs or xfs (though I haven't seen any real speeddifference between any of the 3 ) but it is (In my experience) much more reliable.

Reiserfs is very good, very fast with small files, but has the occasional oops (esp. in combination with NFS or tar) that crashes the system (hard) and/or buggers up the filesystem completely. StarOffice (5.2) install always kernel paniced on one of my boxes on the reiserfs , but not on an identical box (also with reiser)

I still run it on my home box (which isn't networked) with no complaints.

I haven;t played around with XFS much (and certainly not with any of the latest versions) but found it too unwieldy and not stable enough for my tastes.

I wanted to throw in my experiences here. I've used reiserfs for over a year without any problems, both with NFS (and tar -- how could tar cause a problem?). I also just completed my own rather unscientific tests of the speed of ext2/ext3/reiserfs/JFS all on SuSE 7.3 Pro. I wrote a tiny C program to read/write files which I then tested on an 8 MB partition each time formatted clean with a different file system. The files were read/copied from this partition which should have increased any performance differences between them. I copied files ranging in size from 342 bytes to 133560320 bytes (entire kernel source archive). I ran each copy mumtiple times and averaged them. I used the time program to print stats on each run. While this was hardly scientific, it did give me some idea of the real world performance. Of course, all tests run on the same PC/hardware and same configuration other than the one test partition. No optimization was done on any of the file systems and the default block sizes (from YaST2) were used. Here is what I observed from fastest to slowest: JFS EXT3 EXT2 ReiserFS The biggest difference on the 133560320 copy -- JFS (avg 49.780s) to ReiserFS (avg 54.066s), an 8.61% difference. Now, this does not agree with other articles I've read that say EXT2 should be fastest since it does not journal. I can't explain why the numbers came out this way. Even with reiserfs the slowest, I have found it's performance to be fine and it has been quite stable, even with NFS. The JFS site states that you should NOT use NFS on a JFS partition, but I have not tried it. Finally, I have not used ext3 or JFS for any length of time, so I can't give firm opinions about them, and I haven't done anything at all with XFS yet. YMMV. Best Regards, Keith -- LPIC-2, MCSE, N+ wielder of vi(m), an ancient, dangerous and powerful magic free your mind, and your OS will follow

I'm looking to facilitate a box for database operations (MySQL?) and have heard that XFS was the journalling fs of choice for large file management. Thanks, Anthony

On Tue, Feb 26, 2002 at 07:33:12PM +0100, Gerhard den Hollander wrote:

...

* Derek Fountain <fountai@hursley.ibm.com> (Tue, Feb 26, 2002 at 04:07:39PM +0000)

...

On Tuesday 26 February 2002 2:46 pm, you wrote:

...

I had to compile my own 2.4.10 kernel to get XFS support under SuSE 7.3. Does the 2.4.16 SuSE kernel support XFS?

I would recommened ext3 as a journalling fs. It might not be as fast as reiserfs or xfs (though I haven't seen any real speeddifference between any of the 3 ) but it is (In my experience) much more reliable.

Reiserfs is very good, very fast with small files, but has the occasional oops (esp. in combination with NFS or tar) that crashes the system (hard) and/or buggers up the filesystem completely. StarOffice (5.2) install always kernel paniced on one of my boxes on the reiserfs , but not on an identical box (also with reiser)

I still run it on my home box (which isn't networked) with no complaints.

I haven;t played around with XFS much (and certainly not with any of the latest versions) but found it too unwieldy and not stable enough for my tastes.

I wanted to throw in my experiences here. I've used reiserfs for over a year without any problems, both with NFS (and tar -- how could tar cause a problem?). I also just completed my own rather unscientific tests of the speed of ext2/ext3/reiserfs/JFS all on SuSE 7.3 Pro.

I wrote a tiny C program to read/write files which I then tested on an 8 MB partition each time formatted clean with a different file system. The files were read/copied from this partition which should have increased any performance differences between them. I copied files ranging in size from 342 bytes to 133560320 bytes (entire kernel source archive). I ran each copy mumtiple times and averaged them. I used the time program to print stats on each run. While this was hardly scientific, it did give me some idea of the real world performance. Of course, all tests run on the same PC/hardware and same configuration other than the one test partition. No optimization was done on any of the file systems and the default block sizes (from YaST2) were used.

Here is what I observed from fastest to slowest:

JFS EXT3 EXT2 ReiserFS

The biggest difference on the 133560320 copy -- JFS (avg 49.780s) to ReiserFS (avg 54.066s), an 8.61% difference. Now, this does not agree with other articles I've read that say EXT2 should be fastest since it does not journal. I can't explain why the numbers came out this way.

Even with reiserfs the slowest, I have found it's performance to be fine and it has been quite stable, even with NFS. The JFS site states that you should NOT use NFS on a JFS partition, but I have not tried it.

Finally, I have not used ext3 or JFS for any length of time, so I can't give firm opinions about them, and I haven't done anything at all with XFS yet.

YMMV.

Best Regards, Keith

Anders Johansson wrote:

On Tuesday 26 February 2002 20:32, Keith Winston wrote:

...

Here is what I observed from fastest to slowest:

JFS EXT3 EXT2 ReiserFS

I can believe that JFS won. I'm currently running JFS on LVM and I've noted a marked performance increase compared to reiser that I was running before the upgrade.

I'm a little more doubtful about ext3 beating ext2. I mean, ext3 *is* ext2 only with added journalling. It's remarkable that an *added* workload would improve performance.

I just read an article last night about a test (I can't find the original URL now but a search turned up http://www.quest-pipelines.com/newsletter-v2/linux2.htm which is a reprint) to compare the real impact on the various 'cooked' linux file systems, and actually ext3 DID win the tests. BTW, this was the recommended for a database server. -- Joe & Sesil Morris New Tribes Mission Email Address: Joe_Morris@ntm.org Web Address: www.mydestiny.net/~joe_morris Registered Linux user 231871

On 26 Feb 2002, Keith Winston wrote:

Here is what I observed from fastest to slowest:

JFS EXT3 EXT2 ReiserFS

The biggest difference on the 133560320 copy -- JFS (avg 49.780s) to ReiserFS (avg 54.066s), an 8.61% difference. Now, this does not agree with other articles I've read that say EXT2 should be fastest since it does not journal. I can't explain why the numbers came out this way.

Your results are rather interesting because I have reiserfs specifically because of its journaling and its speed compared to ext2fs. I have a nice, old, slow hard drive on my machine and a nice, kickass fast SCSI Plextor CDRW drive. Ripping CDDA can be as fast as 30X [i.e., ~4.4MiB/s] and ext2 simply can't keep up, just like vfat. reiserfs has no problems, however. This was in the days of (the horribly unstable) early 2.4 kernels with a shitty VM, so my results may be irrelevant at this point, but I haven't looked back after switching to reiserfs. As always, there is no FASTEST, but rather, best for a specific task. I have found a good set of bencharks (of course, IMO) for ext2, reiserfs, and XFS [http://bulmalug.net/body.phtml?nIdNoticia=642] . The link is for the Engish translation (original in Spanish). The benchmarks include various different kinds of operations, including copying, deleting, and random seeking. Unfortunately, ext3 was not mature enough at the time of the benchmarks. -- Karol Pietrzak PGP KeyID: 3A1446A0

...

I wanted to throw in my experiences here. I've used reiserfs for over a year without any problems, both with NFS (and tar -- how could tar cause a problem?).

I use amanda for backups, amanda uses gnutar using gnutar to backup the / filesystem will oops the kernel with reiserfs-related panic.

What version of the kernel and reiserfs rpms were you using? Is this ancient history? Chris Shaker

On Tuesday 26 February 2002 6:33 pm, you wrote:

I haven;t played around with XFS much (and certainly not with any of the latest versions) but found it too unwieldy and not stable enough for my tastes.

I installed XFS some time ago and thought it was great. Fast and stable with lots of features like ACL, separate disk journalling and other clever tricks. None of which I'd ever use. :) I stopped using it because SGI didn't keep up with stable kernels (I roll my own). This was especially important in the early 2.4.x phases when patches to 2.4.x were arriving every couple of weeks. Linus would release, say, 2.4.6, and I'd be stuck on 2.4.3 because the updates to XFS weren't available. I switched to ReiserFS. Just recently, however, I've noticed a problem with ReiserFS and OpenAFS. When the AFS cache is on a ReiserFS based partion, and the cache needs flushing, the AFS code oopses. Might be time to try JFS instead... -- 7:58am up 13 days, 23:39, 1 user, load average: 0.04, 0.05, 0.09

Any feedback from NAMESYS (ReiserFS) people?

No, I've gone down the AFS route to try to find out what's happening. Nothing so far...

I, too, am concerned about XFS support and am leaning towards ResiserFS, however, I want large file support for a DB Server. I'm thinking that I might be able to live with ReiserFS since version 4.x will, hopefully, out in Sept'2002.

I think the XFS guys are better on the ball now, and as the 2.4.x kernel has settled down quite a lot you might find XFS is the best answer if you're happy to recompile your own kernel. I wouldn't base too many hopes on a Sept'2002 deadline for ReiserFS features you want. It's bound to be late, and even if it's not, it'll take time to stablise... I can't remember if the JFS version released as open source supports large files - I think it does. If I'm right that might be your best bet. JFS is fast, reliable, and is very actively maintained. -- 2:23pm up 14 days, 6:04, 1 user, load average: 0.35, 0.15, 0.05