HP CIFS Client A.02.02.03 Administrator's Guide

lookupStrategy As you probably know, the HP CIFS Client maps between
NFS requests and SMB/CIFS requests. On the NFS side,
files are referenced by unique identifiers, called NFS file
handles. On the HP CIFS side, files are referenced simply
by their path. The HP CIFS Client must be able to determine
the path given to an NFS file handle. There are two
strategies available to do this:
• pseudoInode
This strategy derives the NFS file handle as a hash
value from the path. The hash is chosen in a way that
makes efficient lookups possible, as long as the depth
of the file in the directory hierarchy is lower than 27.
The advantage of this strategy is the low memory
consumption: Files can be looked up on demand,
nothing has to be stored. The main disadvantage is
that NFS file handles change when files are renamed.
This leads to a conflict with Unix semantics when open
files are renamed: After renaming, the handle of the
open file is stale and the file can not be accessed
without reopening. It also conflicts with a bug in the
caching code of the Solaris NFS client where the
writeback occurs only after closing the file, not during
closing the file.
• database
In this strategy all NFS file handle to file path relations
are stored in an internal database. This is the most
secure and most compatible approach. The
disadvantage is that all this information must be kept
in memory. The HP CIFS Client needs about 500kB
more real memory and about 10MB more virtual
memory for each share that uses this strategy.
The database strategy is the default.
nfsTimeout This integer variable defines the initial timeout in 1/10
seconds that is used by the kernel when it requests data
from HP CIFS Client. This value is doubled on each retry.
Together with nfsRetransmit, this defines the absolute timeout
for NFS requests. A value of 50 (5 seconds) avoids frequent
retries of already running (slow) requests and ensures a total
timeout of about 2 minutes. This should be sufficient even
for the slowest devices and links. If you use a jukebox, it
may also be necessary to increase requestTimeout.
nfsRetransmit This integer variable defines the number of retries the kernel
attempts when HP CIFS Client does not reply in time. The
timeout starts with nfsTimeout and is doubled on each retry.
Retransmissions should not be necessary, because HP CIFS
Client should not lose any requests. However, if your
system's NFS client puts high loads on NFS servers and has
small maximum socket buffer sizes, requests can get lost
due to buffer overflows. A value of 5 (which is also the
default) should be a good choice. You may want to
experiment with nfsTimeout to get the optimum performance
even with frequent buffer overflows.
Configuration Parameters 53