Re: PowerTop discourages beagle


On 5/19/07, Debajyoti Bera <dbera web gmail com> wrote:
I dont have the required kernel, but I straced an idle beagled (idle = built
with --disable-xss, ran with  --disable-scheduler; this disables any
crawling/indexing but query is allowed, and does not do any periodic
screensaver check).

Beagled is perfectly idle in this setup. Err... near-perfectly idle. Beagle
uses a thread which wakes up every 30 sec to check battery status
(reading /proc/ac_adapter/... and stuff). That's the only one working and
diligently waking up at about 30 secs. Rest of the time, beagled is waiting
on a poll().

Arjan (one of the powertop guys) just popped into IRC and mentioned
that when he ran strace on his running beagled that it was waking up
more than just 10 times a second.  On his system, those 10/sec
accounted for one third of all the wakeups on his laptop with a full
desktop environment running.  He wasn't running it with
--disable-scheduler -- which may point to the scheduler as the most
likely culprit -- but he was running with the disable-while-on-battery
setting on.

Remember, when stracing, always pass in -f so that all threads and
child processes are accounted for.

Anyway, the important thing here is not to be reactionary and
defensive against PowerTop, which is a sweet little tool.  Yeah, even
under ideal circumstances Beagle will quite likely be a major consumer
of CPU power; the data that PowerTop provides will allow us to (a) fix
obviously broken stuff that we never would have seen before because a
tool like this didn't exist and (b) have access to useful data which
will allow us to make better informed policy decisions (like whether
to run the indexer on battery by default).


[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]