From: Wally E. <Wal...@ut...> - 2003-09-12 13:51:25
|
Joshua Aune wrote: >Check /proc/sys/net/core/[rmem*,wmem*] and friends on the nodes. If I >remember correctly the init script for nfs in rh8 fiddles with these for >the better. Also see http://tldp.org/HOWTO/NFS-HOWTO/performance.html >for some ideas and reccomended values for these settings. Thanks for the tip. I didn't know that those files were there. That wasn't the problem, however. Those files were the same on both systems. To fix my NFS slowness, I added bg and soft options to the clustermatic mounts in fstab: bg,soft,nolock,intr,rsize=8192,wsize=8192 0 0 That got my speed to about 8 seconds for a 250MB transfer, which is fine for now. I will be using the tips from you and Michael Madore to tweak it some more. Thanks for your help. At 12:52 PM 9/11/2003 -0600, Joshua Aune wrote: >Doing this on one system took me from 6MB/s to 70MB/s with a single >server and single client communicating over gigE :). This required tuning >on both the client and server settings. > >The system that hit 70MB/s on an iozone throughput test had the >following options: rw,rsize=8182,wsize=8192,noac > > > And a sample from /etc/fstab on a non-clustermatic node with the same > > hardware: > > nfsserver:/exports/home /home nfs bg,soft,intr,rsize=8192,wsize=8192 0 0 > >Hope this helps, >Josh |