Hi,
I compared local disk, NFS and GlusterFS with bonnie++ on my old nodes
(Athlon 1 GHz), and seeks/s was 43 seeks/s compared to ~180 for local disk
and ordinary NFS. I guess FUSE etc have overhead compared to a a kernel
solution.
For large file transfers both NFS and GlusterFS saturated a fast ethernet
network. When using 200 Mbit/s NFS scaled up, but not really GlusterFS. I
do however believe that this is a limitation processors but I'm not really
sure.
I will do some tests the following week with nodes with P4 2.8 GHz and
gigabit. Here are the old results anyway:
http://www.update.uu.se/~jerker/tmp/glusterfs.bonnie.txt
(The main good thing with GlusterFS is of course that many servers may be
used in parallell so even if a single server has slightly more overhead
than a kernel solution with NFS, more servers may be added and the files
be distributed among them.)
However, I havn't tried MogileFS yet. :)
Regards,
Jerker Nyberg.
On Fri, 9 Nov 2007, Ask Bjørn Hansen wrote:
Hi,
Anyone here have a deployment of GlusterFS? I have a client who's having a
hard time getting their application updated to use MogileFS to store and get
files. GlusterFS, if it works, would let them keep their filesystem mounted
storage while getting some of the benefits of Mogile.
Any thoughts?
- ask
--
http://develooper.com/ - http://askask.com/
