1. Deployment Planning Guide
  2. NFS Recommendation for Siebel File System

NFS Recommendation for Siebel File System

The Siebel File System (SiebFS) is typically installed on a central global host that shares it to connected clients. These connected clients could be Siebel servers, dedicated ("thick" or remote) clients, a Siebel document server, and so on, depending on the version and configuration.

To ensure that a file accessed by one client can't simultaneously be opened by another, the file system's internal lock mechanism is used. Therefore the host file server itself must implement a global locking concept. This file locking is implemented differently among various operating systems.

Note: This recommendation applies to all types of shared file systems, particularly when the server host is connected using a Network File System (NFS).

Siebel administrators and network architects who wish to successfully implement the Siebel File System, must consider the configuration prerequisites and implementation best practices when the host server is connected with NFS.

If the file server is hosted on Windows using the Server Message Block (SMB) protocol, then file locking is enabled by default and no extra steps need to be taken.

With UNIX and Linux, Samba can provide multi-platform SMB access for Siebel servers running on Windows.

However, in a pure UNIX deployment, the remote share is usually implemented using NFS and certain configuration steps should be verified.

The NFS servers must have the lockd and statd daemons enabled, in addition to basic NFS daemons that implement mounting and accessing a share. These locking daemons must be tuned for the number of threads on various platforms in order to manage the high volume of concurrent lock requests a large-scale Siebel system generates.

NFS server on AIX:

# chssys -s rpc.lockd -a 511

# stopsrc -s rpc.lockd; startsrc -s rpc.lockd

It's recommended that lockd be enabled on every client machine for better load distribution.

NFS server on Solaris or HP-UX 11i v3 (Itanium):

/usr/lib/nfs/lockd [nthreads] nthreads should be set to a value of 200 initially.

This can also be set by defining the LOCKD_SERVERS parameter in the nfs file. The lockd thread tuning for HP-UX is only available with 11i v3 (Itanium). Other HP-UX versions are unsuitable as SiebFS hosts. The local lock mount option (llock) isn't supported by Siebel, because the SiebFS relies on global, system-wide file locking. Local locking causes data inconsistencies.

Note: On Solaris 11.x or above, sharectl utility can be used to set/retrieve LOCKD_SERVERS parameter. See System Administration Commands for Solaris.

NFS server on Linux:

For Linux deployments, it's strongly recommended that only the NFS v4 protocol is used. NFS v4 implements a more robust lock subsystem and therefore should be used whenever possible.
Note: Contact your supplier should a different file sharing protocol such as GPFS, SAN, NAS, and so on, be used, to ensure a file locking concept equal to the above is available.

IMPORTANT NOTICE: To reduce the load on the locking subsystem, the Anonymous user's preference file should be set to read-only. Accessing read-only files doesn't generate a write lock request, since the file won't get altered. In other words, no writes are generated during the anonymous phase of the sign in process. The anonymous user's preference file is accessed for each session sign in. Failure to set it read-only significantly increases the number of file lock requests. For example, to set the preference file for user "GUESTCST" to read-only, run the following command: cd filesystem/userpref chmod a-w "GUESTCST&Siebel Universal Agent.spf".

The read-only setting will remain established when the corresponding user account is dedicated solely to anonymous sessions. If best practices are ignored and the account is also used for regular session log-in, the preferences file will be updated and the file attributes reset to read-write.

In the case where the anonymous user preference file no longer exists in the userpref folder, it will be automatically recreated with read-write attributes, in which case the write attribute must be manually removed again.
Note: If locking isn't implemented correctly, large numbers of 0-byte files with UNIX UTC timestamps could be found in the Siebel Files System root directory, as noted in Empty Files Are Generated in Siebel File System (KB430985). Some components don't require User Preferences functionality and might have the CFGSharedModeUsersDir set to local disk storage to reduce dependency on NFS, thus reducing the likelihood of finding hanging/locking issues related to .spf files residing on NFS.
Below is a list of some components that might have this change implemented:
  • EAIAnonObjMgr_enu
  • EAIObjMgr_enu
  • WfProcBatchMgr
  • WfProcMgr

Example: change param CFGSharedModeUsersDir=/local/path/userpref for comp (or compdef) WfProcBatchMgr

In addition to the above, SavePreferences preferences can be set to False to avoid saving User Preferences at the component task completion. Example:change param SavePreferences=False for comp (or compdef) WfProcBatchMgr. In certain custom implementations, SavePreferences might be required.

It's recommended that you perform testing in a lower environment to verify all works as expected before going live on Production.