From: Sergio Gelato (Sergio.Gelato@astro.su.se)
Date: Thu Sep 04 2003 - 05:49:12 EDT
Just applied the latest PK4 to one of my V4.0G boxes, and some Solaris (7, 8)
NFS clients started having trouble with NFS over TCP. The symptoms are that
the RPC calls reach the server (verified with tcpdump on the server side)
but elicit absolutely no RPC response from Tru64. The TCP packets are
promptly ACKed, but "netstat -f inet" shows them remaining in the receive
queue. The server is essentially idle, and quite responsive to everything
else (including other NFS clients).
This affects specific clients (their identities are persistent across
reboots) which are otherwise identical (hardware and software) to other
unaffected clients. So far none of our Linux or Tru64 clients have been
affected; this may be purely a Solaris-Tru64 incompatibiilty.
An apparently (so far) robust workaround is to force the use of UDP for
NFS mounts from V4.0G PK4.
Questions:
1) Has anyone else seen this?
2) Are there better workarounds? Kernel tuneables I should check?
3) Which of the 192 patches I just applied as part of PK4 is responsible?
Complicating factors, hopefully irrelevant but...:
a) I'm running AFS (OpenAFS 1.2.9) on this machine. That's the only
non-HP kernel module in there.
b) One of the local filesystems is AdvFS.
I'll summarize.
This archive was generated by hypermail 2.1.7 : Sat Apr 12 2008 - 10:49:34 EDT