Desktop Filesystem Benchmarks in 2.6.3
mfedyk at matchmail.com
Sun Mar 7 05:00:20 UTC 2004
Ivan Ivanov wrote:
> I don't think that XFS is a desktop filesystem at all.
> This is from XFS FAQ:
> Q: Why do I see binary NULLS in some files after recovery when I
> unplugged the power?
> If it hurts don't do that!
> * NOTE: XFS 1.1 and kernels => 2.4.18 has the asynchronous delete path
> which means that you will see a lot less of these problems. If you still
> have not updated to the 1.1 release or later, now would be a good time!
> Basically this is normal behavior. XFS journals metadata updates, not
> data updates. After a crash you are supposed to get a consistent
> filesystem which looks like the state sometime shortly before the crash,
> NOT what the in memory image looked like the instant before the crash.
> Since XFS does not write data out to disk immediately unless you tell it
> to with fsync or an O_SYNC open (the same is true of other filesystems),
> you are looking at an inode which was flushed out to disk, but for which
> the data was never flushed to disk. You will find that the inode is not
> taking any disk space since all it has is a size, there are no disk
> blocks allocated for it yet.
> This same will apply to other metadata only journaling filesystems. The
> current linux kernel VM will write out the metadata after 1/60th of a
> second and the data after 30 seconds. So the possibility of losing data
> when unplugging the power within 30 seconds is quite large. The only way
> of being sure that your data will get to the disk is using fsync in the
> program of sync after closing the program.
> I am trying XFS from 2.4.6 and I can reproduce this case easy. Simply
> write some file and unplug the power during write. And we are talking
> for desktop :). XFS is the worst case for recovery too.
> For desktop filesystem speed is not mandatory. Hard disk speed is most
> importatnt. And much more important is recovery. So I think that ext3 is
> the best solution for desktop system. It performs well and is the most
> recovarable linux filesystem.
Thanks to Suse, reiserfs v3 will have an ordered mode like ext3, and
reiser4 does data journaling by default.
Until the other filesytems write out the data before the inodes are
journaled (ie ordered mode), they are not suitable in an environment
where the power can go out unexpectedly.
More information about the Ext3-users