ONCplus B.11.31.15 Release Notes (5992-5685, September 2012)

8. Problem
• The write operation may be interrupted when very large files are copied over NFS.
• System crash may occur under heavy load conditions.
Solution
This problem has been fixed in ONCplus B.11.31.14.
Workaround
For ONCplus B.11.31.13, the workaround is to use the forcedirectio option for the NFS
mount in case of loopback mount or cross mounts with very less UFC cache (say 1%) configured
on NFS client systems.
Problem
There is a possibility for an HP-UX 11i v3 system to experience a system hang under the
following conditions:
• An application writes a huge NFS file to a filesystem using NFS loopback mount.
• Two systems NFS-mount file systems from each other and simultaneously write huge
amounts of data to each other.
Causes
On an HP-UX 11i v3 system with NFS loopback mount enabled, there is a possibility for an
application to consume the entire file cache memory while writing a huge NFS file. This could
lead to a deadlock between the NFS client and NFS server modules.
For example, when the NFS client consumes the entire file cache, the server cannot execute
any of the WRITE requests sent by the client because of the non-availability of file cache
memory. This can result in non-availability of file cache memory for other mounted file systems
on the node.
There is a high possibility for this problem to occur if the system is configured with low file
cache memory.
For example, when the filecache_min and filecache_max parameters are configured in the
range of 1% to 5% of the total physical memory.
Solution
Starting from ONCplus B.11.31.15 onwards, a fix has been implemented for both NFS v3
and v4 client versions.
To avoid this issue, the tunables must be set to the following values:
• kctune nfs_enable_write_behind = 1
• kctune nfs_enable_ufc_threshold = 1
• kctune nfs_ufc_threshold_percentage = 50
For more information about the new tunables, see the Managing NFS and KRPC Kernel
Configurations in HP-UX 11i v3 whitepaper
NOTE: All NFS filesystems have to be unmounted if there is any change in these tunables
or UFC filecache_max. Please note in this version a new tunable ‘nfs_ufc_threshold_percentage’
is introduced to cater for both v3 and v4 mounts instead of old tunable
nfs3_ufc_threshold_percentage. From ONCplus B 11.31.15 version,
nfs3_ufc_threshold_percentage is made dummy tunable. Start using new tunable
‘nfs_ufc_threshold_percentage’ instead of ‘nfs3_ufc_threshold_percentage’.
Installation Information 19