[lug] journaling fs

[Ian S. Nelson]

llornkcor wrote:

> Anybody here using a journaling fs regularly? Any impressions you
> might have on it? How about the migration to using it? Are you
> using it on the / partition?

Here's my big compound reply..

I've had a Reiserfs partition on one of my home partitions for a while now.
Only in the 2.x days did I ever lose any data or have any crashes.  Now I've
got about 80GB in Reiserfs at home, no problems at all.  At work  I've got
about 40GB in it and I haven't lost any data there either.  I've been testing
it and XFS extensively because I'm going to use some kind of log structured
filesystem in a product I'm building.   Mandrake  supports ReiserFS out of
the box these days too..

The only real odd behaviors I've noticed with ReiserFS is that there is some
obscure bug in it where performance is CPU bound,  it's much more noticable
on slower machines (pentium 1 class machines) where it can use as much as 5
times more CPU cycles as Ext2 doing mapped and unmapped I/O.  Hans knows
about it but doesn't know where the  bug is yet.  Also, on power offs, files
that are currently be written to can get filled with garbage.  I haven't lost
data but I've had extra trash added to them, the ReiserFS team thinks there
is some odd bug where DMAs keep going a tiny bit after power is cut to the
processor, I kind of think that is BS because I haven't seen it on other
filesystems and it is regular on Reiserfs.  Either way it's pretty minor.
I've been debugging boot stuff and kernel hacking and I've tested the "fsck"
a ton, I generally only press reset on one of my machines at work and I
haven't had any problems at all.  Fsks 15G in about 7 seconds and pops right
back up.

I run it on root partitions but I keep and ext2 /boot partition, just because
I don't fully trust lilo yet and I'm not an expert at grub.

To migrate to it, I recommend tar.  Tar up one partition and then untar it on
to another.

Overall I'm quite pleased with ReiserFS and XFS too.  Both are very
sophisticated modern filesystems and I think both are suitable for day to day
use and maybe even production in a limited capacity.

Alan Robertson wrote:

> Using ReiserFS for everything but /boot (can't easily).
>
> The issues include:
>         dump/restore doesn't work
>         kernel NFS daemon doesn't work

I think it does in the newer cuts.


>         Can't use Tom's Root Boot, and others
>         There is no reliable fsck, so the journalling better work right

I haven't had fsck problems, yet.. ;)

I have also made a Tom's disk that will work with ReiserFS.  it is fairly
easy get the Tom's build area working, this is kind of screwy, you have to be
root on my machines to make it work and I almost never use root for
development..  Then you can kill the modules out of the second "disk"  I
believe they are in toms...blah./usr/lib/modules instead of
toms...blah/lib/modules   anyways, you can wipe them all out.  Then in the
first directory kill the kernel and replace it with a new one. I've made a
disk that uses a 2.4.0test5 kernel with ReiserFS applied.    The trick is to
pick the hardware you *need* and then compile it all in to the kernel.  Mine
is under 800K.    To pick up a little bit of loose change, delete the man
pages too.  I got mine to fit and had about 15K to spare.  I may have also
killed "vi"    My kernel had IDE, SCSI, ReiserFS and a 3com ethernet in it,
it was functional enough to rescue but probably not as good as a real Tom's.
The rescue disk was what I wanted though.    If you want I can dig out my
build area and pin down exactly what I did to make it work.    I also started
exploring a "net Tom's" disk that I could plug in, have it load up and then
pull a root off of a server somewhere, it hasn't amounted to much more than
notes scribbled in my notebook at this point.

If you use a kernel newer than 2.3.35 you can also look at big drives, this
might be a more important reason to do it.  >32G drives are getting popular.

Sean Reifschneider wrote:

> The journaling is done such that as long as you cleanly un-mount the
> journaled file-system, you can re-mount it as ext2.  The btree directory
> entries can be enabled on a directory-by-directory basis, so you can try
> before you buy there as well.  That was last I heard though, things may
> have changed.

I don't believe that there is btree in ext2/3 yet.  I'm pretty sure it
isn't.  On the devfs list and Reiserfs list Stephen seems to be very
supportive of using Reiser as the FS of the future when Hans isn't making an
ass of himself.  I don't want to go too quote crazy but reiserfs does have an
expander in some kind of state (alpha? I assume) I've not used it but there
isn't and probably won't be a shrinker very soon.  The expander may be
intended for LVM, it's in the utils directory with recent reiserfs patches.
They are building a repacker, I'm not positive what it is yet but you can
optimize file layout based upon access, journal location, and drive geometry,
I would assume that if you repacked a partition then it would be pretty easy
to shrink it but that's just a guess.

IBM's JFS seems to be making pretty good progress too.  There is almost no
community support for it though, I don't know why that is, the xfs and
reiserfs lists get 10-80 messages a day.  Steve Best, an IBM engineer is the
only who seems to post to JFS.  It could be a dark horse.



--
Ian S. Nelson                                           __o
Nelson_ at attglobal.net.NOSPAM                            \<,
                                                      ()/ ()