Figure 1: Implementing node-specific files within the CFS name space

Node A

/var/opt/hpws/logs

 

 

 

 

k

 

 

 

in

 

 

 

l

 

 

 

 

c

 

 

 

 

i

 

 

 

 

l

 

 

 

 

o

 

 

 

b

 

 

m

 

 

y

 

 

 

S

 

 

 

 

CFS storage

Node B

 

/cfsmnt01

 

apache

logs htdocs

--> /var/opt/apache/logs

Symboliclink

/var/opt/hpws/logs

 

•

Symbolic link is created once on CFS

 

Local root disk

•

Node-specific log files are stored in the local

Local root disk

 

 

file system (which is not part of CFS)

 

Cluster-wide files are best placed on CFS. For example, application data files usually belong to these kinds of files on CFS. Other cluster-wide files (like application configuration files) could be stored locally on node-specific file systems but would need to be kept in sync manually. Since those files change frequently, storing them on CFS is probably the better choice.

For application executables, the decision is often made in favor of local storage because they are updated less frequently than configuration files. Storing application executables locally on each node also increases the application availability slightly. Independent software vendor (ISV) applications might provide installation utilities that consider these points and provide cluster-wide installation options.

Synchronizing data updates on CFS

Multi-instance applications that access the shared data on CFS from multiple nodes concurrently must synchronize their write access to the data. All three CFS bundles provide a single file system schema that is cache-coherent and Portable Operating System Interface for UNIX® (POSIX)-compliant, which means that different processes running on different nodes of the cluster can access the CFS concurrently.

The lockf system call is available for applications to synchronize concurrent access of data within a file between multiple processes. With CFS, this lockf functionality extends from processes on a single node to processes on all cluster nodes that have a CFS mounted. Those locks are advisory only according to the POSIX standard. The HP-UX proprietary S_ENFMT option to enforce locking is not supported on a CFS. With lockf, a range of a file can be locked with the granularity of a page (greater than or equal to 4 Kb) to increase parallel write access to the same file.

14