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

nfs performance tuning for hp-ux 11.0 and 11i systems page 18

Notes:

Page 18July 22, 2002

local

filesystems

Filesystem Recommendations (part 1)

• Use VxFS filesystems whenever possible

• Use block size of 8KB if possible, otherwise use 1KB

• When using VxFS 3.3, tune with vxtunefs(1M),

otherwise use VxFS 3.1

• Specify 16MB logfile size via “mkfs –o logsize”

i.e. 2048 (8KB block size) or 16348 (1KB block size)

VxFS filesystems are recommended over HFS filesystems on both NFS clients and

servers. VxFS filesystems have a huge advantage over their HFS counterparts

because of the way VxFS interacts with buffer cache. VxFS filesystems track their

buffer cache usage on a file-by-file basis, which allows the kernel to invalidate

VxFS files from the cache much faster than HFS.

An 8KB block size should be used if possible. While this might result in wasted

space if the filesystem is comprised mainly of very small files, the 8KB block size

allows VxFS to perform “block” reads when servicing NFS read requests.

If you wish to use VxFS 3.3 in your environment, you should familiarize yourself

with the vxtunefs(1M) command. In many situations, an un-tuned VxFS 3.3

filesystem will perform much worse than a similar VxFS 3.1 filesystem.

When building the VxFS filesystems on your NFS server, the maximum intent log

size (16MB) should be specified, as this allows the largest number of requests to be

pending in the logfile before any intent log maintenance is required by the kernel.