Hi,

As background, I have a CentOS 5 workstation accessing data on an NFS server in a different internal domain (VLAN?) with traffic passing through multiple firewalls, and the "disk" access is painful. One job is taking 4x longer, just because the data is on the NFS server instead of local disk.

However, we have other CentOS 4 workstations that are in a different domain, and they do not show the same slowdown.

So, I am thinking the problem is either the firewall latency, slow switch for my internal domain, bad match between ethernet driver and local hardware that's too new, or bad NFS client configuration.

My question is, "How would you troubleshoot this networking problem and converge on the problematic link?"

Thanks!

NFS uses the RPC daemon and is not reliable over the busy network. It has the tendency of getting hanged up and is pretty slow. Try to use samba instead. Export the same share through samba and check if the performance improves.

I have had issues with NFS over WAN and hence have replaced all the NFS mounts to Samba. However, NFS works nicely on LAN. Try increasing the rsize and wsize options for NFS and also the timeout value.

Hope this helps!!
Rahul Khare

I cannot remember it quite clearly, but something was said that NFS works better in congestion if one uses Udp instead of Tcp. Something about tcp getting hung up on retransmissions. Sorry I could not be clearer.

Hope it helps

Lazlow

Even with UDP I have faced issues. The problem lies with portmapper which is used to assign ports for NFS data transfer dynamically. You can try udp instead but I dont think it would rectify the issue.

Rahul Khare

I have NFS locked down to specific ports. Part of FC5 NFS setup. I could not get selinux,iptables and NFS happy any other way