Re: ext2 -> reiserfs
Hall Stevenson wrote:
> Anyone with opinions on ext3 vs reiserfs, IBM's journaling fs, or others
> ?? This one sounds the "easiest"... Also, how soon might this be
> implemented in Linus' kernel instead of requiring a patch ??
I've been reading up a bit on jounaling filesystems lately. Here's what
I have learned:
1. Most journaling fs's currently journal only the metadata (information
about inode allocation and so on -- internal structural information
about the state of the filesystem), but do NOT actually journal the
contents of files. This means that despite the journaling, it IS
possible to lose some data if you have, say, a loss of power (although
it is NOT possible to actually corrupt the filesystem itself). I believe
only ext3 journals the file content data. The ext3 developers claim this
is not only better protection for your files, but also a performance
boost, since all synchronous file writes are made to the journaling file
only, and written out to the "real" files asynchronously.
2. Another plus for ext3 is that it is fully backward- and
forward-compatible with ext2. You can convert ext2 to ext3 in place
(once you have a kernel that supports ext3) just by creating the journal
file. An ext3 filesystem can always be mounted as ext2, as well, which
is useful for rescue diskette boots or downgrading (if you decide, for
some reason, to stop using ext3).
3. ReiserFS handles very small files (< 1k) by storing the file data
with the directory entry rather than in a separate place (as most
filesystems do, and as ReiserFS does for larger files). This makes
accessing those files very, very fast. It can also save space by
combining the tails of files, i.e. the last sector of file A may also
contain the last sector of file B if the combined size of the two tails
will fit into a single sector. This is a space-vs-speed tradeoff; you
can disable tail sharing if you prefer speed.
4. SGI's XFS is not recommended for the typical home desktop system. It
makes much greater demands on the system than ext3, ReiserFS, or IBM
JFS. It comes into its own on multi-processor file servers.
5. I see no particular reason to use IBM JFS. It does not seem to have
the speed of ReiserFS, the SMP benefits of XFS, or the ext2-compatiblity
of ext3. My reading has not turned up any particular pluses of JFS that
make it a strong competitor to the others. If anyone knows better, please
let us all know!
I have never actually used a journaling filesystem, but I'm going to try
ext3 with my 2.4.7 kernel and see how it goes.
Craig
Reply to: