We have a couple of HP-UX 11.0 systems running Oracle 8i (8.1.6). Each of these systems has about 10-15 small to medium sized Oracle databases on the system. In order to separate tables, logs, and archives for availability and recovery in the event of a filer volume failure, we are using 3 volumes per database. We are using qtrees allow for the mounting the database tables, logs, and archives separately which allows us to use failover between the HP servers for each database. This all boils down to needing 45+ NFS mounts per client.

HP still uses separate biod processes (the default is 4 biod processes) , and I think each running biod can handle 2 concurrent I/O requests (this may or may not be configurable by through some kernel parameter - not sure what the parameter is though).

Does anyone know how to tune the HP-UX client for the best performance for this many mounts? I currently think we should bump up the number of biod processes to about 32. We will probably bump up to 16 for a while just as an intermediate test before we go to 32. Does anyone know of other changes that will increase NFS client performance?

Rick Hulsey
Southwest Airlines
2425 Wyman
Mail Stop 4DC
Dallas, Tx., 75235
rick.hulsey@wnco.com
(214) 792-7188 Office
(972) 880-6882 Cell
(800) 915-3747 Pager