[NTLUG:Discuss] Time to switch away from ReiserFS?

[Robert Pearson]

On Thu, Jul 10, 2008 at 10:35 PM, Chris Cox <cjcox at acm.org> wrote:
> Leroy Tennison wrote:
> ...
>>
>> Hope this doesn't open a can of worms but, what is it about reiserfs
>> that makes it suitable for true enterprise deployment where no other
>> file system is?
>
> It's about the only filesystem that can be dynamically resized
> online in EVERY situation (and QUICKLY).  Very fast filesystem
> creates even on very large filesytems.  Handles directories with
> tons of files very well.  Doesn't fragment much (most I've ever
> seen is about 9%... I've seen double that with ext3).
>
> XFS is usually growable... but can't be shrunk (not that you
> need to shrink a filesystem that often... just a nit).  There
> was a fairly recent bug involving stack sizes in the kernel,
> XFS and LVM2.... I think those issues have been resolved.
>
> We've been running two NAS'd home directory areas of 800G
> each on top of reiserfs for about 7 years.  We have other
> reiserfs filesystems of even greater size used in our
> disk to disk backup, some of which might be several years
> old (but in general, a lot of that changed as we moved
> to different storage devices).
>
> I have ext3 filesystems that exhibit behavior that is
> UNEXPLAINED.  Yes... they are trashed.  In particular
> it was a RHEL 4 machine.  There are directories where
> you can't create files at a certain depth or lower... but
> you can create files in other trees with equal depth
> on the exact same filesystem (bizarre).  Filesystem
> passes fsck.
>
>
>
>>
>> While we are on the subject, from what I've heard ext3 journals both
>> metadata and data while the other file systems don't.  Why is this?
>
> It's an option... and it's an option on reiserfs on any relatively
> modern distro (data=ordered (default, same as ext3 default), data=writeback
> (just metadata like old reiserfs), data=journal (full data journaling)).
>
> There was a time when reiserfs did only metadata journaling and
> it wasn't ordered journaling... that one was fixed a few years back.
> The journal (full data journaling) option is relatively
> new.
>
> Remember that enterprise commercial journaled filesystems often just
> do metadata journaling (like old resierfs, XFS and JFS).
>

It's kind of lonely over here in the "Waiting for ZFS" line. Anybody
want to join?

XFS was the finest (and fastest point to point) file system I had ever
seen in the late 1990's. SGI was on a power roll.
Then the bottom fell out. I don't expect XFS to ever be a top player
again. Hasta la vista, baby.
Then in 1998 at the Mass Storage Conference I saw GFS (Global File
System) co-developed by Sandia and ???(SGI).
The first Storage aware and centric file system. It suffered a fate
worse than death.
I had hoped for a GFS resurgence when Red Hat bought it from Sistina.
Not to happen. Apparently it is a "point product" solution in the Red
hat arsenal. Good approach. File systems are "point product" solutions
rather than "one size fits all" as they are commonly used.

I stopped using ReiserFS when openSUSE no longer had it as the
default. I've had problems with all of them in Linux and I am not a
power user. I had never used ext2 or ext3 until then. Now everything I
have is ext3 and sometimes I have to fight to get the OS to install
that. Damndest thing I ever saw.  It keeps defaulting to ext2? I have
heard encouraging comments about ext4. We'll see...
Actually I am trying to learn how to use LVM and ext3 or just LVM.
I am very LVM challenged. I don't find it intuitive at all. People
look at me funny when I say that. My reason for saying that is that
Fedora once, during an install when I requested LVM, made the whole
disk a PV and then it either created multiple VGs or multiple LVs in
lieu of partitions. I had to low-level format that drive with the
manufacturers tool to reuse it. Neither Linux "fdisk" (reported no
partitions) nor LVM (reported no PVs, VGs or LVs) could do a thing
with it. Fedora would install to it.
Ah! Sweet mysteries of life.

Robert