[plug] JFS pros, cons?

[Bernd Felsche]

On Fri, Apr 04, 2003 at 09:14:46PM +0800, Denis Brown wrote:
> On Fri, 4 Apr 2003, Bernd Felsche wrote:

> > On Thu, Apr 03, 2003 at 02:39:04PM +0800, Denis Brown wrote:
> > > I'm going through the motions of setting up what I hope will be a
> > > high availability Linux machine.   It will have a UPS, dual SCSI
> > >
>  
> > What? You don't rsync to a warm standby host over gigabit? :-)

> Would love to if I could find a generous benefactor :-)  In fact
> in the long term I hope to have a couple of these boxes at
> separate sites and I was thinking to link them using CODA.  Rsync
> sounds like a better idea.  But that is In The Future (tm)

Does CODA do compression?

Problem with rsync is that it has to do its checksums... if the
deltas are integral to the filesystem, then the compressed deltas
can be sent over a piece of wet string.

The problem I've seen in the past is when the network-based mirrors
broke; they took forever to catch up, and were sometime in danger of
stalling the master system indefinitely... not a pretty sight.
Guarranteed to happen just as you're about to have that important
dinner with somebody special...

Frankly; I'd like to see network-mirroring at the LVM level.  Yep;
scary. But it does allow you to put the preferred filesystem on top
without complication of filesystems. Just wondering how one can
patch in a "physical volume" - maybe the much-abused SCSI layer on
top of a network-based "scsi adapter".

Stop me please; I'm likely to have a nightmare if I keep going like
this: Somebody else can have that nightmare. My kernel hacking
skills are exhausted by "make xconfig".

> > > For example, is anyone using JFS and have comments to make
> > > regarding it?   I'm not so much fussed about data loss caused by
> > 
> > AFAICT, JFS is still a bit of an experiment.

> Hmmm... I get that impression too.

> > If you need journaling, then Reiserfs is "more proven". ext3 is nicely
> > robust. Reiserfs has the performance advantage on loarge directories
> > and small files. ext3 works well on large files.

> One of my tasks for this server is to support databases and web-based
> calendar and resource services so in the main I'm probably looking at
> large-ish files from few directories.   Being able to manage the ext3 fs
> with ext2 tools is probably an advantage for me, not that I've had to use
> ext2 tools in anger many times -- touch wood.   Also it seems that ext3
> offers data as well as metadata journalling, but I have yet to read more
> on Reiser; it may well do, too.

ext3 appears to work better than Reiser for large files with many
updates on them. Reiser seems to get "overwhelmed" with the re-write
activity, but shines on big directories and conserves a great deal
of space when there are lots of small files on the filesystem.

> Having to investigate the various options is rewarding in itself (for me
> at least).   I can see why Reiser went for B trees and dynamic allocation.
> I'd value a clarification on what you mean by Reiserfs being "more
> proven."   I don't yet know enough about the history and use of ext3 and

It's more proven than JFS... Reiserfs has been in SuSE Linux for
4 years or so. It's been "standard" in that distro for about 2 years
so there are "millions" of test-bunnies. I'm one of the
test-bunnies; so are 4 of my customers.

> Reiserfs to understand what you might mean.   I know that some people have
> mentioned data corruption issues with laptops, when using Reiser, but an
> article by Daniel Robbins (Gentoo CEO) claims that is due to lt hdd
> hardware issues.

Indeed. IIRC, it's to do with the power-saving curses^Wfeatures that
resulted in Reiserfs' journals not being completely written.

> > It's hard to find one filesystem that fits all requirements. I've
> > been looking for 20 years. :-)

> Hear, hear!   Many moons ago I wrote a fs for a Z80 system, that was
> collecting brain (EEG) data in great slabs.   Even though you might think
> the data arrays would have a fixed size, because of the experimental
> conditions, this was not always true.   In those days, every byte was
> precious A$1000 for a 10 megabyte (yes, megabyte) hdd.   Many schemes
> tried in an effort to find the magic bullet.   It finally came -- the
> PC -- with more storage than we could dream about: 70 megabytes.   "You'd
> never fill that!"

Luxury. I had 10 Unix users on 45 megabytes of hard disk.

> Cheers and thanks to all for replies so far.   I'm going to spend some
> time in the literature but the way things are looking atm, ext3 has the
> edge.

If you're not creating/destroying files frequently, it seems to win.
But be prepared to use different filesystems throughout the system.

-- 
/"\ Bernd Felsche - Innovative Reckoning, Perth, Western Australia
\ /  ASCII ribbon campaign | I'm a .signature virus!
 X   against HTML mail     | Copy me into your ~/.signature
/ \  and postings          | to help me spread!