HP Tru64 UNIX and TruCluster Server Version 5.1.B-4 Patch Summary and Release Notes (13156)

For example, ORACLE RAC uses a cluster interconnect for its own lock manager traffic and
an increase in latency could lead to unpredictable results.
Additionally, multi-instance applications can increase the burden on the cluster interconnect
and affect inter-member cluster communication so should follow the configuration
recommendations in the next section.
• An Enhanced Distance Cluster should not be used for the following workloads due to their
I/O characteristics and directory locations. (Mail and print usually keep their data stores in
the system directories):
— • NFS server
— • Mail server
— • Print server
A.3 Enhanced Distance Cluster Recommendations
Although Enhanced Distance Cluster configurations provide access from both sites to all shared
data storage in the system, the volume of traffic to off-site storage, either through the storage or
cluster inter-site connections, should be minimized to maintain performance levels. Therefore,
you should be aware of the following recommendations are applicable. Not following these
recommendations can have detrimental effect on system performance and availability.
The main purpose of most of these recommendations is to design a configuration that minimizes
I/O latency and contention issues within both the cluster and the san storage inter-site connections:
• The configuration of an application and its data storage should be located within the same
site. Doing otherwise can cause significant performance degradation to the application’s
runtime performance.
• Each application should have its associated data in separate file systems, or AdvFS domains,
in order to locate the CFS server for these file systems with the application.
• Applications should not use any storage/directory within the system tree (for example, root,
/usr, /tmp, /var, /var/tmp, and so on). Instead they should use dedicated file systems
that can be co-located with the application. If the application requires a directory in the
system tree, a symbolic link should be used to redirect these directories to a separate
storage/file system.
• The use of remote I/O via DRD to serve devices is strongly discouraged due to the increased
traffic on the cluster interconnect possibly resulting in unacceptable latencies.
• Although the use of LSM for striping is supported, it is not recommended for use.
• Although data replication is supported between HSG80 arrays, their use as shared storage
within an Enhanced Distance Cluster it is not recommended. This is due to the inter-site
distance limitations of the HSG80's remote data replication, which would significantly impact
the Enhanced Distance Cluster configuration.
• The use of a cluster alias for network-client access is discouraged due to the potential for
additional traffic on the cluster interconnect if the alias router is not located with the
application.
If a cluster alias is used, each application must have its own cluster alias. Each application
should have its cluster alias router located on the node where the application is executing.
This can be achieved by using the cluamgr command and adjusting the values of selection
weight and routing priority. During a failover of the application, the cluster alias also should
be relocated to the same node.
A.3 Enhanced Distance Cluster Recommendations 185