NFS Performance Tuning for HP-UX 11.0 and 11i Systems

nfs performance tuning for hp-ux 11.0 and 11i systems page 35

Notes:

Page 35July 22, 2002

Why not just launch hundreds of

biods and be done with it?

• 11.0 client’s NFS read() and write() paths use the global filesystem

semaphore to protect key kernel data structures and I/O operations

• Acquiring the filesystem semaphore effectively locks out all other

filesystem related operations on the system – not just other NFS

requests but requests for all filesystems (VxFS, HFS, CDFS, etc.).

• The overhead involved with contending for and acquiring the

filesystem semaphore becomes detrimental to NFS and general

filesystem performance when many biods run on 11.0 clients.

biod

Difference between 11.0 and 11i

Filesystem semaphore contention drastically reduced in 11i

Based upon the earlier description of what the biods do, one might conclude that

the more biods running the better your NFS client will perform – so why not just

launch 200 biods and be done with it?

On HP-UX 11.0 NFS clients this is not a good idea because the client’s read() and

write() paths use the global filesystem semaphore to protect many kernel data

structures and NFS I/O operations. What this means is that whenever a biod

processes a read or write request it must acquire the filesystem semaphore, which

effectively locks out all other filesystem related operations on the system – not just

other NFS requests but requests for all filesystems (i.e. VxFS, HFS, NFS, CDFS,

etc.). Therefore, if an 11.0 NFS client runs a large number of biod daemons, the

overhead involved with contending for and acquiring the filesystem semaphore

becomes detrimental to NFS and general filesystem performance.

This filesystem semaphore contention issue is drastically reduced in 11i, which

means that an 11i client could potentially benefit from running more biods than an

11.0 client.