HP-UX 11i September 2001 Release Notes

New and Changed Disk and File Management Features
Network File System Support for TCP/IP (new at 11i original release)
Chapter 9
146
Network File System Support for TCP/IP (new at 11i
original release)
Network File System (NFS) is now supported over the connection-oriented protocol,
TCP/IP for NFS versions 2 and 3, in addition to running over User Datagram Protocol
(UDP). TCP transport increases dependability on wide-area networks (WANs).
Generally, packets are successfully delivered more consistently because TCP provides
congestion control and error recovery.
As a result, with this new functionality, NFS is now supported over WANs. As long as
TCP is supported on the WAN, then NFS is supported also.
The mount_nfs command now supports a proto= option on the command line where the
value for proto can be either UDP or TCP. (In the past, this option was ignored.) This
change allows the administrator to specify which transport protocol they wish to use
when mounting a remote file system.
If the proto= option is not specified, by default, NFS will attempt a TCP connection. If
that fails, it will then try a UDP connection. Thus, by default, you will begin using TCP
instead of UDP for NFS traffic when you begin using the 11i version of HP-UX. This
should have little impact you. You do, however, have the option to specify either UDP or
TCP connections.
If you specify a proto= option, only the specified protocol will be attempted. If the server
does not support the specified protocol, the mount will fail.
nfsd now opens TCP transport endpoints to receive incoming TCP requests. For TCP,
the nfsd is multi-threaded. For UDP, the nfsd is still multi-processed.
Kernel TCP threads execute under the process nfskdtcp. When counting the number of
nfsd processes, keep in mind the following algorithm: An equal number of nfsd's that
support UDP will be created per processor and only one nfsd that supports TCP will be
created. In the case of a four-way machine and NUM_NFSDS=14 (set in
/etc/rc.config.d/nfsconf), 17 nfsd’s will be created: 16 for UDP (4 per processor)
and 1 for TCP.
nfsstat will now report TCP RPC statistics for both client and server. The TCP
statistics will be under the connection-oriented tag and the UDP statistics will be under
the connectionless-oriented tag.
AutoFS will support the proto= option in the Automounter maps and will have the same
behavior described above under the mount_nfs command. In the past, this was an
invalid option.
Automounter will not support NFS over TCP.
Unlike the 11.0 patch release of NFS over TCP, there is no enablement flag for NFS over
TCP. By default, NFS will attempt to use TCP.
The kernel RPC layer was modified to support TCP connections over NFS. A new
streams module, rpcmod, was added to manage the TCP connections. These changes are
internal to the NFS implementation and are not user accessible.