svc: server socket destroy delayed

Norman Bridges n.w.bridges at
Fri Jan 10 10:50:00 CST 2003


Judging by my google search this may be a golden oldie, but I can't pin
down the problem ...

We're running a number of P/E 2500s (1GB memory, dual Intel Pro/1000
gigabit adapters) as file servers offering nfs and samba. The nfs load
is mostly light.

Software is Redhat 7.3 just patched with all outstanding errata patches
including kernel 2.4.18-19.7.x. Each system immediately started
reporting intermittent messages

 svc: server socket destroy delayed

I've since modified the nfsd startup so that it does (from

  echo 262143 > /proc/sys/net/ipv4/tcp_rmem
  echo 262143 > /proc/sys/net/ipv4/tcp_wmem
  daemon rpc.nfsd $RPCNFSDCOUNT
  echo 131071 > /proc/sys/net/ipv4/tcp_rmem
  echo 131071 > /proc/sys/net/ipv4/tcp_wmem

This was suggested in some nfs performance tips. All nfs clients are
using nfsvers=3,rsize=8192,wsize=8192.

I think the number of "server socket destroyed" messages is reduced, but
they are still happening.

Ideas anyone?

Thanks in advance.

Norman Bridges

N.W.Bridges at[tel] +44 1782 583073:[fax] +44 1782 584234
Keele University:Information Services:Keele:Newcastle,Staffs:ST5 5BG:UK

More information about the Linux-PowerEdge mailing list