IBM Hub/Switch, Installation Manual

Page: 431 / 590

Chapter 7 HPSS User Interface Configuration
HPSS Installation Guide September 2002 431
Release 4.5, Revision 2
7.4 NFS Daemon Configuration
Before the HPSS NFS daemon can be started, any existing AIX or Solaris native NFS daemons
must be stopped and prevented from restarting. This is important because the NFS protocol does
not provide a way for clients to specify which of two daemons is wanted. When the system is set
up correctly, there should be no 'nfsd' or 'mountd' processes running. If nfsd was running
before, it may be necessary to free up the ports it allocated by using the 'rpcinfo -d' command. Otherwise the
HPSS daemon may not start correctly. Also be sure to stop rpc.lockd and rpc.stad. The HPSS NFS daemon
does not support file locking.
Additional HPSS NFS configuration information is specified through the HPSS exports file and
through environment variables. If the default server name for the HPSS LS is not used, the
environment variable for the LS, HPSS_LS_NAME should be changed in the hpss_env file.
Similarly, if the default for the NFS Server descriptive name is not used, the environment variable
HPSS_NFS_DESC_NAME should also be changed in the hpss_env file.
NFS allows you to traverse junctions, but be aware that certain functions may not behave as
expected after a junction traversal (i.e. “
cd ..
”). By default, NFS junction traversal is disabled. To
enable it, add
HPSS_NFS_ENABLE_JUNCTIONS = “on”
to the hpss_env file before bringing up
the Startup Daemon. This environment setting will also allow junctions to be mounted (i.e. “
mount
hostname:/junction
”).
HPSS NFS does not support yellow pages (Sun MicroSystems’ Network Information Services) to
validate hosts. HPSS NFS does provide an option to validate the network address of hosts
attempting to mount HPSS directories. The default configuration disables this check. To enable this
check, define the variable HPSS_MOUNTD_IPCHECK in the hpss_env file.
Note: For users who have several NFS clients concurrently updating and reading the HPSS
namespace and depend on having their namespace changes immediately reflected in all clients, it
is necessary to perform NFS mounts with the noac option. Here is an example:
mount -orw,intr,timeo=30,noac hpssserver:/hpss /hpss
Using the noac option will degrade performance since it prohibits caching. Only use it if NFS clients
rely on immediate HPSS namespace updates.
If people are using NFS to mount file systems from several hosts, a situation can develop where one
NFS daemon gets bogged down and then users can't access the other file systems either. This is a
common problem with all NFS file systems and is not peculiar to HPSS. However because the HPSS
NFS daemon is prone to bogging down, it is especially likely to affect other file systems.
To avoid this problem, sites may want to advise their users on the best way to set up their NFS
mount points. All of the mount points for a particular NFS host should be kept in a directory
separate from the other hosts. (This is probably a good idea even if the clients are not mounting
HPSS files.) One convention for doing this is to name the mount points by the name of the host that
is exporting the file system. For example, if a host named tardis is exporting HPSS directories /
home and /public, and a host named jupiter is exporting the Unix directory /home, the mount
points might be: /nfs/tardis/home, /nfs/tardis/public, and /nfs/jupiter/home
If these names aren't considered user friendly, the site can use symlinks to establish friendlier
names. For example, /hpss could point to /nfs/tardis, and /users could point to /nfs/jupiter/home.