Re: NFS tuning?
Am Samstag, 11. Dezember 2004 17:03 schrieb Ralf Gesel|ensetter:
> as I suspected NFS to be a brake in performance, I looked around for
> information on how to tune NFS.
Here, I got a collection of interesting URLs:
http://www.linuxquestions.org/questions/showthread.php?s=&postid=960123
http://www.unixcircle.com/features/nfs.php
There:
Tuning NFS
Some suggestions to tune client and server for performance:
1. If clients and servers are not on the same LAN or the network is
overloaded, use TCP if possible.
2. Increase read/write packet data size. Need to increase read/write
value if experienced some error messages such as ``fragments dropped
due to timeout'' while still mounting.
3. Augment the number of NFS server daemons on a busy server.
4. Bump up the number of kernel threads to serve asynchronous NFS I/O
requests on the NFS client.
LINUX:
1. Use UDP. Only switch to TCP if client/server is not on the same LAN
or packets get dropped often.
2. Edit /etc/fstab and on the 4th field which is the filesystem mount
options, add rsize=32768,wsize=32768 after rw.
3. The default NFS daemon is 8. Crank up to a high number by
editting /etc/init.d/nfs and change 8:
# Number of servers to be started by default
RPCNFSDCOUNT=8
to a suitable number.
4. Not sure if Linux has this feature to tune.
Some hints for diagosing are on
http://billharlan.com/pub/papers/NFS_for_clusters.html
Regards
Ralf
Reply to: