I think that counting clock cycles is a necessary but not sufficient improvement in the metrics.
I did extensive research into alternative methods for determining capacity/headroom in virtualized environments in 2002, published the basic idea in some papers (Capacity Planning for N1/Virtualized Datacenters or somesuch) and wrote up the most detailed form as a patent which was filed in 2004/5 after I left Sun. I'm sure someone at Sun can find the patent and make some use of it. Adrian On 9/18/07, Keith Bierman <[EMAIL PROTECTED]> wrote: > > On Sep 14, 2007, at 5:03 PM, Walter Bays wrote: > > > Good point. Speaking of which, as the clock frequency changes with > > power > > management events, should the CPU utilization change along with it? > > What about completely asynchronous designs? > > I guess I'm more of the Snead school of thought, the concept of "CPU > utilization" sans some workload seems to be a dated notion of > increasingly less utility. > > > Keith H. Bierman [EMAIL PROTECTED] | [EMAIL PROTECTED] > Strategic Engagement Team | AIM: kbiermank > <speaking for myself, not Sun*> Copyright 2007 > > > > > _______________________________________________ perf-discuss mailing list perf-discuss@opensolaris.org
