Programmatically obtaining kernel metric info..

[David Gourley]

Does anyone know any mechanism to obtain some of the more interesting Glance style metrics programmatically?

For example, I know that data equivalent to many of the metrics can be obtained from the pstat* APIs. However there are a number of really interesting metrics which I can't see corresponding pstat APIs to, e.g. GBL_PRI_QUEUE, GBL_CPU_CSWITCH_UTIL (and a number of others related to queue size on particular kernel resources). From reading HP-UX 11i Tuning and Performance, it appears that there are a number of non-idle activities which may show up as idle time with the simpler APIs - is there any way to find out how much *useful* CPU time is used without loading Glance?

There are issues with Glance's footprint for the particular problem I'm investigating.

[Steven E. Protter]

I am attaching some sar based data collection scripts. They may work for you.

SEP

[Mark Greene_1]

"is there any way to find out how much *useful* CPU time is used without loading Glance"

Try top -f [filename] -d [count]

where "filename" is where you want the output saved, and "count" is the number of iterations.

mark

[David Gourley]

A bit more investigation: top / ps / sar all use the pstat APIs. Glance appears to use the kernel instrumentation APIs defined in ki.h (but documentation on these appears to be sketchy at best).