This topic describes how to plan for organizing the global-namespace
by partitioning it into filesets.
At the root of the global namespace is the global fileset. The name
of the global fileset is always ROOT. The directory name of the global fileset
is specified when you set up the global namespace, for example
as sanfs.
When you create a new fileset, you attach it to a specific location in
the global namespace, creating a hierarchy. You can attach the fileset to
the global fileset or to another to another fileset. When a fileset is attached
to another fileset, it is called a
nested fileset.
Tip: In the SAN File System environment, you can create regular,
non-fileset subdirectories only from the client machines. As a result, the
metadata servers cannot recreated directory structures that contain a mix
of filesets within subdirectories and regular subdirectories. To simplify
disaster recovery, attach filesets only to the global fileset (root directory),
not to regular subdirectories under another fileset. The mkdrfile command
output can then be used to completely restore the top of the global namespace
tree before using the client-based backup application to restore the rest
of the global namespace.
These are some guidelines for partitioning the global namespace into filesets:
- Keep data used by an application in the same fileset. Do not to split
the data across multiple filesets.
- If you must split the data used by an application across multiple filesets,
observe natural divisions in the application data to minimize cross-access.
- Create enough filesets so that workload can be easily balanced. It is
easier to balance the workload with 10 filesets per server rather than two
filesets per server.
- Define filesets so that fileset quotas is meaningful.
- Keep the number of file objects per fileset under 100 000 to 1 000 000.
- Keep number of fileset transactions per second for a single fileset under
1 000.