Eric Saxe wrote:

> Alexander Kolbasov wrote:
> > Bill Holler <Bill.Holler at sun.com> wrote:
> >
> >
> >     Bill> It will be useful to have c-state information observable 
> > (including latency
> >     Bill> and power savings) during development and testing.  Should c-state
> >     Bill> information be in a kstat such as the cpu_info kstat?
> >
> > Isn't c-state information extremely transient? I am wondering what value
> > can you get if it is placed in kstats. IMO something like total time in
> > various states might be more useful (if it is possible to obtain).
> >   
> PowerTop conveys the total time in various c-states (over a given 
> sampling interval). I'm not sure how useful the "total time since the 
> system  booted" would be, since that would probably include periods of 
> time where the system was busy.

You can easily use it to monitor current time in c-states by substracting 
accumulated time between snapshot - similar to the way various utilities deal 
with micro-state times which are also accumulated totals. And keeping 
accumulated totals allows us to easily get totals over last period T.

- akolb


Reply via email to