Application Use Cases for the HP Serviceguard Storage Management Suite, November 2009

HP Serviceguard Storage Management Suite bundles with

CVM and CFS

The T2775xx and T2777xx HP Serviceguard Storage Management Suite bundles include CVM and

CFS. T2777xx also include Oracle-specific tools and database accelerators. These three bundles will

support clusters of up to sixteen nodes.

CFS-specific application integration considerations

CFS offers new functionality to applications but provides new challenges in integrating them with CFS

and exploiting the features delivered with CFS. The topics discussed in the next three sections apply to

multi-instance active/active applications using a CFS:

 Comparing node-specific and cluster-wide files

 Synchronizing data updates on CFS

 Performance and scaling with CFS

The fourth and fifth sections discuss topics important for single-instance failover applications using a CFS:

 CFS primary placement for failover applications

 Shutdown for single-instance failover applications in CFS clusters

Comparing node-specific and cluster-wide files

One of the first steps in integrating an applic

ation with CFS is to decide which files are node-specific

and which are cluster-wide (shared among all nodes). This task is application-dependent. For some

files, you may have a choice between storing them locally or on CFS. For others (like the data files for

a database), the location is predetermined—it is definitely on the shared file system.

An easy way to deal with node-specific files is to place them on the private file system of a local node.

Sometimes, applications expect a node-specific file at a certain path location. Apache, for instance,

expects the error_log file at logs/error_log, relative to the ServerRoot directory. If the

ServerRoot directory has been configured on CFS to share the configuration files, a symbolic link

must be used for the logs directory to point from the shared CFS location to a local file system (in

–s/var/opt/hpws/logs/cfsmnt/apache/logs). Otherwise, all nodes will write to the same

error_log file, and the administrator will not be able to identify which message came from which server.