Hmmn, the question of busy/waiting/not-busy for a CPU could be
adressed in rather the same way as was done with disks, with
two orthogonal metrics, one for the percent of the time there
was work in the queue for the CPU, and the other for the queue
length when there was work present, where fractional lengths
correspond to the traditional "% busy".

See the diagram at line 594 in /usr/include/sys/kstat.h, where
the are treated as a Riemann sum.

The area under the curve illustrated there is mathematically
convenient, and matches the intuitive idea of work done, making
it easy to plot and interpret visually.

--dave (Hi, Jim!) c-b
 
 
--
This messages posted from opensolaris.org

Reply via email to