On Sun, Nov 12, 2006 at 04:32:27PM +0100, Federico Giannici wrote: > Now that the DNS problem is solved, it SEEMS that the problem with NFS > is reduced.
Interesting... let me know what else you find out. > 1) Is NFS activity in some way related to DNS? Not really. Well, both go through the network. :-) > Anyway it could be that my "mail delivery" program depends on DNS, and > so it delayed the delivery... Possibly. > 2) Is it possible that, if a certain number of processes are already > using NFS, subsequent attempts by other processes is STOPPED, until some > other process RELEASE the use of NFS??? Not the use of NFS, but of certain resources. It's like that all over the kernel. > If this is true, what I can do to increase the number of concurrent > processes writing via NFS? > I increased "vfs.nfs.iothreads" to 20, but it seems not be enough... Well, it would be nice to have real locking for NFS, so some polling constructions could be avoided. But given the amount of people interested in helping, that's too far of a goal. So yeah, try bumping that for now, if you feel the server (and the network) are comfortable with the load. -p.
