[NBLUG/talk] Recommendations for journalling filesystems?

Scott Doty scott at corp.sonic.net
Tue Aug 1 13:53:40 PDT 2006


On Tue, Aug 01, 2006 at 10:42:20AM -0700, Augie Schwer wrote:
> On 7/27/06, Lincoln Peters <petersl at sonoma.edu> wrote:
> >I now know that ReiserFS is not a practical option for a journalling
> >filesystem, if for no other reason due to its inadequate recovery tools (I
> >don't know enough about filesystems to debate any other technical merits it
> >may or may not have).  So I ask my fellow NBLUG members: what journalling
> >filesystems have you used, and which (if any) would you recommend?
> 
> "Benchmarking Filesystems Part II" from January's Linux Gazette:
> 
> http://linuxgazette.net/122/piszcz.html

One filesystem that I did some work with over the 4th-of-July weekend was
Red Hat's "Global File System", or GFS.

I have some gndb vs. nfs benchmarks that I did on July 3rd:

   http://ponzo.net/hob/bench/

gndb is the system for sharing Red Hat's GFS cluster file system:

   http://www.redhat.com/magazine/021jul06/features/gfs_update/

* The "triple" benchmarks are bonnie running on all 3 machines on the "hob"
cluster -- {a,b}.hob using gndb to the GFS filesystem, which was local to
c.hob.

* gndb shares used GULM -- lock server was running on c.hob.

* Each system has two 3 GHz Xeon 64-bit dual-core
  CPU's (i.e., "quadcore") w/1GB of RAM

* The nfs mount was using:
   vers=3,proto=udp,rsize=16384,wsize=16384,rw,nodev,intr,bg

...which is the optimum settings that we came up with back in 2001 or so
(YMMV).

(Note that modern Linux kernels now default to "sync" for their nfs mounts,
unless you specify "async".)

( BTW:  GFS and the Red Hat cluster tools are in Fedora Core 5 ... :)

 -Scott

p.s. http://www.redhat.com/magazine/021jul06/features/gfs_update/



More information about the talk mailing list