DragonFly BSD
DragonFly users List (threaded) for 2005-04
[Date Prev][Date Next]  [Thread Prev][Thread Next]  [Date Index][Thread Index]

Re: dragonfly pdf documentation

From: Andrew Hacking <ahacking@xxxxxxxxxxxxxxx>
Date: Fri, 01 Apr 2005 17:43:37 +1000

FWIW: In recent testing with later versions of the 2.4 Linux kernel line
I found that [in _my_ circumstances], EXT-3 was far more reliable than
JFS.  After repeated weekend long tests on a system with a hardware
watchdog that would drop the power randomly while a lot of
file-creation/deletion was being performed, only EXT-3 survived without
corruption. JFS on the other hand managed to get its meta-data screwed
up resulting in files with shared data blocks, and contents of old
files.  There are a number of journaling modes in EXT-3, I chose the
least performant but most reliable mode which provides both meta-data
and file-data journaling.

I think my JFS experience has more to do with its integration into the
linux kernel than with the JFS itself.  A look at recent 2.4 kernel
change-logs shows a number of JFS related fixes and interestingly a hell
of a lot of XFS fixes.  If bug fixes are anything to go by, EXT-3 fixes
seem rare in comparison indicating that it may be more stable.


On Fri, 2005-04-01 at 09:35, Miguel Filipe wrote:
> Hi all,
> On Thu, 31 Mar 2005 15:46:10 +1000, Steven Shaw <steshaw@xxxxxxxxx> wrote:
> > On Wed, 30 Mar 2005 23:53:13 +0800, Bill Hacker <wbh@xxxxxxxxxxxxx> wrote:
> > > Boot from XFS, use JFS for non-boot storage in the
> > > places it serves better than XFS, and you have the
> > > foundations for as durable a Linux as can be.
> > 
> > My laptop used a single large XFS "boot" parition for everything when
> > I was on Debian. Now  I'm on Gentoo, I use ReiserFS for the same
> > purpose. After this discussion my new laptop will be running XFS again
> > :-).
> Watch out... don't take every word here said has pure truth.
> I've used:
> - UFS+logging (solaris 7,8,9)
> - ext3
> - reiserfs
> - xfs
> - ufs2 bsd style, with and without softupdates.
> - apple UFS, HFS+ with and without journaling
> From my experience he worst of them, in terms of reliabilty I would
> say ext3..  I'm talking / partitions on student laboratories with CS
> students from year 1 to year 5. These 100 machines take quite a
> beating... and 10% of them at the end of the semester have the windows
> partition fucked up. The linux partition.. on the elder ones do get
> screwed.. like 2,3 machines out of 100.
> all the other files systems.. only on server rooms + personal machines
> (ibook laptop runs linux on ext3 and stood up quite some kernel hangs
> (my bad... bad kernel module of mine) and power failures) .. From
> reiserfs & XFS I can only say good things.. and they are in ftp
> servers, rsync servers, mail servers, web+mysql servers.. with a bunch
> of diferent workloads , and there have been some powerfailures outside
> the server room. The only total filesystem loss I had was with ext3 +
> nv driver caused system crash.
> So, I refuse to accept that linux FS are worse or more unstable than
> BSDs, or solaris ...
> reiserfs especially, i've used it in so many adverse conditions.. and
> never failed on me beyond recovery. Since linux 2.4.16 reiserfs is
> rock solid AFAIK.
> Also remmember that linux is very popular and takes quite a
> beating/testing from many many people.
> XFS rules for big/medium files, and orthogonally for big file
> systems.. (1/5 TB and up..)
> reiserfs rules for small files
> ext3 ... I use it just because of compatibility (reading my linux / on
> OSX, dumping a linux install on ghost.. etc..) cause elsewere.. it
> sucks..
> On the BSD side, softupdates are good, solaris has journaling, so has
> IBM and SGI, I don't think journaling is a bad concept.. but I don't
> know that much of FS design.
> BSDs do need a fast way to recover 1+ TB filesystems, and performance
> wise, sync mounts SUCK.
> The async/sync battle between BSDs and Linux are a design one.. and If
> one does not aprove the async idead, should allways disable raid and
> disk caches...
> I do approve the async idead, the same way I like raid5 with big
> caches and disks with big caches.
> My knowlegde of BSD is inferior of my knowledge of linux, I'm getting
> my feet wet on linux kernel developement, have used openbsd for 2
> years has my gateway, and 1 year has main machine, 2 years of real
> solaris sysadm work, experienced several times with freebsd, and now
> I'm using dragonflyBSD on a spare laptop and running it on QEMU on top
> of linux.
> So, I don't know as much about BSD FS design as I know about linux FS design..
> BTW, from the design point of view, XFS is the best one, although
> reiser4 has some amazing features, and I'm waiting for it to mature.
> Also, HFS+ is slow has hell.. now it has journaling,  it was slow
> before, it's still slow now..
> But like I said before... take every post/mail/comment with a grain of
> salt, and make your own mind.
> > 
> > Cheers,
> > Steve
> > --
> > Steven Shaw http://c2.com/cgi/wiki?StevenShaw
> > 
> Best regards

[Date Prev][Date Next]  [Thread Prev][Thread Next]  [Date Index][Thread Index]