bismark@alta.jpl.nasa.gov
david_thorburn-gundlach@groton.pfizer.com
kevin@uniq.com.au
putama@ect.enron.com
Simon-Bernard.Drolet@M3iSystems.QC.CA
Suggestions for solving the problem included troubleshooting the NFS network
traffic with snoop, reduce the size of the FDDI MTU to 1500 bytes, and make sure
the 2.6 server is sending NFS V2 over UDP (also via snoop). putama mailed the
Sunsolve NFS and automount FAQs, both of which I'd already read and put aside.
Per the NFS faq, I tried enabling fragmentation on the FDDI interface with
the following command:
ndd -set /dev/ip ip_path_mtu_discovery 0
This didn't seem to make a difference. Snoop showed that the NFS server kept
attempting to retransmit a request for cookie information, which made me think
the NFSD was doing version 3 of NFS.
I ended up killing NFSD and MOUNTD and restarting them with nfsd -a 128 (up
from nfsd -a 64). After restarting NFS, the Sun OS boxes are able to automount
the Solaris 2.6 directories, and snoops appear normal.
My conclusion is that NFSD on the Solaris 2.6 box was not started properly,
and was trying to negotiate NFS version 3 with the Sun OS boxes.
Thanks again for your help,
Jim Dunbar
jdunbar@nyx.net
My original question appears below:
---------------------------------------------------------------------
Dear Sun Managers,
We recently installed Solaris 2.6 on an Ultra-2 Enterprise system with 128 MB
of memory and 150 MB of swap space. The primary purpose of this box is to
provide light NFS services for approximately 250 Solaris 2.5 and 2.5.1 clients.
There are also a few Sun OS 4.1.3_U1 servers that exist in the environment as
"legacy" boxes that will be phased out over the next year. Automounting is the
primary method of attaching to the NFS server.
The NFS server is equipped with a FDDI interface through which the clients
will attach for file services. The clients are attached to 10 mb ethernet
which is switched onto the FDDI ring as needed.
So far, all is well with the Solaris 2.5 and 2.5.1 NFS clients. They see the
server's shared disks and have no problems accessing the data. The Sun OS
boxes are another matter. The automounted disks are shown as mounted, but
when attempting to access the NFS disks, the Sun OS boxes report errors such
as "NFS read failed for server <servername>: RPC: Timed out." These errors
are reported when the NFS server disks are soft-mounted. When changing the
mount type to hard, the Sun OS boxes report that "NFS server <servername> not
responding. Still trying," and never come back to the UNIX prompt.
My question is: Does anyone know of any Solaris 2.6 changes to NFS that would
prevent Sun OS automounts from working correctly? Note that Sun OS standard
mounts work fine. Note also that we have several 2.5.1 servers to which the
Sun OS boxes are doing automounts without a problem.
Thanks in advance for any help you can give me. I'll be sure to summarize.
Jim Dunbar
jdunbar@nyx.net