On Sat, Jun 30, 2001 at 06:25:18PM -0400, Jean-Francois Landry wrote:
>
> On Sat, Jun 30, 2001 at 08:29:04PM +0200, Jens Benecke wrote:
> > I just had a, er, 'lively' discussion with someone claiming ReiserFS
> > is crap because it hogs even the fastest CPU too much, and it
> > uses 4x as much processing power to do metadata operations, and
> > in general is slower because of the journal. My benchmarks don't
> > reflect this, especially on current hardware (ATA-66 and ATA-100
> > disks on VIA chipsets).
it's not that important in many (most?) situations that require fast
file I/O.
the CPU in most I/O bound applications sits idle waiting for the disk.
probably the only place where it would really matter would be in a
beowulf style cluster or rendering farm. it's rare for any application
to do a lot of I/O *and* a lot of computation.
image files are huge, so a render-farm would be better off using XFS
than reiserfs anyway - all benchmarks i've seen (including my own bonnie
runs[1]) show that XFS is faster for large files, while reiserfs is much
faster for lots of little files. overall, reiserfs is faster, but XFS
uses less CPU.
for applications that are primarily CPU-bound, it probably doesn't make
much difference what file-system is used.
for applications that are primarily IO-bound, the best filesystem to use
depends on the nature of the application.
in any case, you have to carefully benchmark all the available
filesystems and weigh up the pros and cons until you've found the one
that works best with your application.
saying that reiserfs is best, or xfs is best, or foo-fs is best, is just
plain silly - every fs has advantages and disadvantages.
e.g. i'd use reiserfs for a mail queue fs or a news spool or Maildir/
style mail spools. for a mail spool with large mbox files, i'd use XFS.
if i had to choose one or the other, i'd probably choose
reiserfs...although it's a difficult choice: reiserfs wins overall on
performance, but XFS has had several years more debugging in real-world
conditions (on SGI's IRIX)
i guess all this is a long way of saying "facts and figures are
relevant, but unsupported opinion counts for nothing".
[1] my preliminary benchmark results are at:
http://siva.taz.net.au/~cas/x-vs-r.html
this is only a benchmark of a single IBM 40GB drive. I bought a pair of
these and had intended to also benchmark a software RAID-0 setup, but
one of the drives was faulty when it arrived (DMA and BAD CRC errors).
i'll do another bonnie run when the replacement arrives.
> OK, all those B* trees suck CPU cycles, and the journal means overhead,
> but in my humble opinion it's worth it. Even with the CPU hogging, most
> IO operations seem snappier to me on Reiserfs (at least compared to
> ext2, can't say about XFS or JFS). And I'm not even talking of the more
> efficient file packing.
>
> You makes your choices, you lives with it ;)
yep.
craig
--
Craig Sanders
Systems Administrator
VICNET- Victoria's Network http://www.vicnet.net.au/
