John, I need more data on your system, I'll send what I need offline.

Thanks,
-Aubrey

John Martin wrote:
> 
> On 04/ 6/11 02:36 AM, Li, Aubrey wrote:
> > poll mode is more lazy to speed down, is this an idle system?(what's
> platform?)
> 
> Idle desktop - metacity, thunderbird, gnome-terminal
> 
> i7 980X + eVGA X58 3X SLI motherboard (latest BIOS)
> 
> > if so, you should wait until the frequency speed down to the lowest one and
> > read the power usage again.
> 
> If I wait long enough it settles to around 5W above event-mode,
> but the power meter jumps around quite a bit.
> 
> > It looks like xcall/pstate transition is gone. I suspect latest scheduler 
> > change
> makes
> > the system more aggressive to race to idle.
> >
> > I attached a dtrace script to capture the idle/non-idle transition number. 
> > This
> is not
> > p/c-state related, and not powertop related. I saw a big regression against
> previous
> > build. Please have a try on your system and let me know the result.
> 
> When in poll-mode:
> 
> # ./idle-state.d
> dtrace: script './idle-state.d' matched 5 probes
> dtrace: 221 dynamic variable drops with non-empty dirty list
> CPU     ID                    FUNCTION:NAME
>    7  75543                       :tick-5sec
>            2425903
> 
>    7  75543                       :tick-5sec
>            2566395
> 
>    7  75543                       :tick-5sec
>            2556586
> 
> dtrace: 339 dynamic variable drops with non-empty dirty list
>    7  75543                       :tick-5sec
>            2350252
> 
> dtrace: 9 dynamic variable drops with non-empty rinsing list
> dtrace: 192 dynamic variable drops with non-empty dirty list
>    7  75543                       :tick-5sec
>            2354720
> 
> dtrace: 52 dynamic variable drops with non-empty dirty list
>    7  75543                       :tick-5sec
>            2565431
> 
> dtrace: 182 dynamic variable drops with non-empty dirty list
>    7  75543                       :tick-5sec
>            2706685

_______________________________________________
pm-discuss mailing list
[email protected]
http://mail.opensolaris.org/mailman/listinfo/pm-discuss

Reply via email to