NFS server not responding ... socket overflows?

[Jeff Pendleton]

From the original post (vmunix: NFS server not responding still trying),

Chris Anoruo wrote

"On the client, use SAM to increase the Mount failure rate(retry), retrans and the timeo before doing at command line prompt /sbin/init.d/nfs.client start I hope you have created the directory."


Thank you for the answer, next i will try to increase the timeo option, as you mentioned.

on my studies i discovered another aspect:

using

netstat -s

on the nfs-server i saw the following output for udp:

udp:
0 incomplete headers
0 bad data length fields
0 bad checksums
25682 socket overflows # !
0 data discards

now i assume that the log-entries in my clients syslog files belong to socket overflows on the server, which belongs to a number of running nfsd's which is too small.

my nfs-server has 64 nfsd's running (configured throug NUM_NFSD switch in nfsd.conf ) .

is it possible that this server-side number is too small for 16 client-workstations running a CAD-client program (IDEAS MS7)?

thanks

[CHRIS_ANORUO]

Use netstat -sr and netstat -sM, if the ouputs are that bad reload ther server.

[Brian M. Fisher · Accepted solution]

NFS client:
Block I/O daemons (biods) facilitate requests to the servers, by buffering requests together. The normal default is 4 biods per client.
If netstat -s -p udp indicates "socket overflows" it should be increased.

NFS Server:
Roughly 3 nfsds should be configured per disk on large multiple-CPU systems. Configuring too many nfsds can leave some of them idle, or it can affect overall system performance.
netstat -s -p udp indicating "socket overflows" would suggest that more nfsds would help.

Brian
<*(((>< er