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

nfs performance tuning for hp-ux 11.0 and 11i systems page 15
Notes:
Page 15July 22, 2002
Copyright 2002 Hewlett- Packard Company
local
filesystems
Measure Filesystem Throughput
Generally speaking, the higher your local filesystem
throughput is, the better your NFS performance will be
Eliminate the NFS layer from consideration (if a filesystem
throughput problem exists it should affect any I/O traffic)
Use tests that don’t require filesystem resources to run
Ø iozone (http://www.iozone.org)
Ø dd(1)
Once the layout of the server’s filesystems has been analyzed and optimized where
possible, the next step in validating your disk subsystems is to measure the
throughput of the client’s and server’s local filesystems. Generally speaking, the
faster your underlying disk subsystems, the better your NFS performance will be.
An approach similar to the network testing described earlier should be used, where
NFS is removed from consideration. If a filesystem throughput problem exists on
the NFS client or server, the problem should affect any I/O traffic, not just NFS.
The testing utilities should allow the administrator to isolate one type of filesystem
I/O at a time (i.e. test read throughput then write throughput) before mixing I/O
types. Also, the tests themselves should not require any filesystem resources to run
as this could affect the throughput results (i.e. don’t test filesystem write
performance with a test that requires reading from a filesystem).
Two of the available tools that meet these requirements are iozone and dd(1).