Re: OT: which journaling file system
On Thu, 19 Jul 2001 00:05, Alexander Clouter wrote:
> > It seems that ReiserFS and XFS are both relatively stable now and
> > not too hard to install cleanly.
>
> XFS has been stable (on IRIX machines) since about 1994. ReiserFS is
> really only a year or two old. So far all the stories I have heard about
> it involve lots of data corruption. Not really very nice :)
There are very few reports of data corruption on the ReiserFS list. Of
recent times a large proportion of the data corruption issues have turned out
to be hardware failings. ReiserFS uses more CPU time than other file systems
and can result in hotter CPUs, machines with defective cooling have been
known to fail from this.
XFS has been widely used on IRIX, not on Linux. To port XFS to Linux they
had to add significant features to the kernel and they needed to re-write
parts of the file system. XFS on Linux is not the same as XFS on IRIX.
> Also another thing to bear in mind is that ReiserFS only journals metadata
> (if I remember correctly) so data corruption is still possible anyway,
> its only a half way mark.
Journalled meta-data means that there is no need for fsck after a power
failure, and it stops cross-linked files etc. If you write a block of data
to end of a file and the power fails then the file will be either extended or
not extended (it's an atomic operation). However the actual data written to
a file may be partially written (if it crosses a page boundary).
Is this such a problem? Do other journalled file systems do it differently?
Early versions of Ext2 journalled all data and were very slow because of it
(it was regarded as a bug).
> As you may of guessed I'm biased towards XFS, mainly because I installed
> it last week on my laptop and everything has been fine (except for a few
You seem very sure of the reliability of XFS after only a week of use...
I've been running ReiserFS on a number of server machines for two years,
although not for the root file systems (due to LILO issues) and some machines
have Ext2 for /home because until recently ReiserFS lacked quota support.
Generally I have found ReiserFS to be quite reliable.
> As for a web article Slashdot (www.slashdot.org) ran an article a couple
> of weeks back. It wasn't a proper benchmark however it gave you an idea
> of performace between xfs/reiser/jfs/ext2/fat32. XFS came out the best in
> all the tests except for massive file deletion. Not really a problem.
ReiserFS delivers the best performance available for small files. If you are
running a web cache (average file size ~13K) then ReiserFS will perform very
well.
Compare XFS and ReiserFS for files <64K in size and expect ReiserFS to win.
> One thing in ReiserFS's favour is that it is good on disk space, I'm not
> sure how good but figures up to about 10% saving have been quoted, this
> however I don't believe.
Why not? ReiserFS by default packs tails of files. This means that if you
create 9 files of 400 bytes they will all fit into a 4K disk block. Ext2
with 1K blocks will take 9K for the same data, Ext2 with 4K blocks (everyone
who has big file systems uses 4K blocks for Ext2 to reduce FSCK times) will
take 36K to store the same data.
I'm sure I could contrive a test to show that ReiserFS saves 99% of the disk
space! ;) But the 10% number being thrown around is probably from some
common file set such as the kernel source.
> I managed to convert my laptop to XFS in about 6 hours. There was no
> usable free space on my 2Gb harddisk so I had to copy across a 10Mbps
> network to a desktop machine. Everything went fine and all the remains is
> / to be done, however I don't see much point in doing it so I probably
> won't bother. Also it allows me to use the normal debian boot disks
> without having to find specially made disk images.
Use initrd and you can load the right kernel module for the file system early
in the boot process. Creating a Debian boot disk that then mounts a regular
rescue disk shouldn't be so difficult.
> One or two other minor issues. NFS and LVM support is fine under XFS (and
> probably JFS too) however ReiserFS bombed out in a nasty way.
How did NFS bomb out for you?
Until recently there were issues with the kernel nfs server and ReiserFS.
Basically if the server was asked for files that weren't in the dentry cache
then it couldn't satisfy the requests. This occurred when the server was
rebooted, or when there was a large number of files open without use so that
memory pressure forced dentries out of cache.
At the time there were these problems I was regularly using kernel NFS on
ReiserFS and it was working fine. My NFS server machines were not under
heavy load, and they had reasonable amounts of RAM (minimum 64M).
The only LVM issue regarding ReiserFS I am aware of is mounting snap-shots
(because ReiserFS tries to replay the journal even when mounting read-only).
I have been using ReiserFS off and on since 1998. I have used it to run mail
servers, web servers, laptops, and NFS-root servers. I have run it on RAID-1
and RAID-0. I have used ReiserFS for testing my hard drive and mail server
benchmark software, these tests have shown deficiencies in ReiserFS
(situations of poor performance), most of which have been fixed. While doing
the same tests I discovered a number of bugs in Linux software RAID support
which caused crashes and data loss (which have also been fixed). Also I
discovered a bug in the Linux NFS client operation which AFAIK has not been
fixed (an NFS server sending bad data after a kernel panic made the client
machine unusable).
--
http://www.coker.com.au/bonnie++/ Bonnie++ hard drive benchmark
http://www.coker.com.au/postal/ Postal SMTP/POP benchmark
http://www.coker.com.au/projects.html Projects I am working on
http://www.coker.com.au/~russell/ My home page
Reply to: