How to set i/o priority for nfs client processes?

The configuration is: A linux server and a nas box (netgear) acting as nfs server.

It is easy for a single process on the linux server to use all i/o bandwidth by simply copying a file from the nfs share to the nfs share. The i/o channel is jammed and all other processes on the server will nearly halt waiting for i/o. Load grows up to 10-20 (four cores), more and more pdflush processes appear... until someone stopps the file copy.

How can I limit the i/o bandwidth the cp process uses? nice will not help of course, but also ionice -c3 has no effect. Does ionice affect nfs mounts at all? Is there something like nfsnice ?

What are your "rsize" and "wsize" values set to?

Normally, modern linux NFS clients negotiate the maximum values with the server, but sometimes, they can end up way off base. For example, we had rsize=1m,wsize=1m in /proc/mounts, not knowing the NAS being unable to support more than 32768. Same slowliness, same effect of load skyrocketing as you describe.

Setting both values down to 32k immediately solved the slowliness and the rising load for us, desktop remained perfectly responsive even while copying gigabytes per NFS. And we have our home directories on NFS...

Perhaps your NAS's NFS server implementation does a little "show off" by offering more size than it can chew...?

Cheers