Re: [patch v8 3/9] sched: set initial value of runnable avg for newforked task
From: Morten Rasmussen
Date: Thu Jun 20 2013 - 06:23:45 EST
On Sat, Jun 15, 2013 at 01:09:12PM +0100, Lei Wen wrote:
> On Fri, Jun 14, 2013 at 9:59 PM, Alex Shi <alex.shi@xxxxxxxxx> wrote:
> > On 06/14/2013 06:02 PM, Lei Wen wrote:
> >>> > enqueue_entity
> >>> > enqueue_entity_load_avg
> >>> >
> >>> > and make forking balancing imbalance since incorrect load_avg_contrib.
> >>> >
> >>> > Further more, Morten Rasmussen notice some tasks were not launched at
> >>> > once after created. So Paul and Peter suggest giving a start value for
> >>> > new task runnable avg time same as sched_slice().
> >> I am confused at this comment, how set slice to runnable avg would change
> >> the behavior of "some tasks were not launched at once after created"?
> > I also don't know the details on Morten's machine. but just guess, there
> > are much tasks on in the run queue. the minimum load avg make the new
> > task wait its time...
> Is there some possibility that since task structure is allocated without being
> set to 0, and it cause the imbalance between runqueues. Then the new forked
> is migrated to other cpus, so that it cause its execution being delayed?
> It is better for Morten to give us more details here. :)
I think Peter's reply pretty much covers it. The problem is when a task
is not running (other task has lower vruntime or blocked for other
reasons) shortly after the task was created. The runnable_avg_period is
very small, so the load_contrib is very sensitive.
Say if a task runs for 1 ms then is blocked for 1 ms and then runs
again, the load_contrib will go from 100% to 50% instantly and then ramp
back up again. So the task load may be quite different from the true
load of the task depending on when you calculate the load_contrib.
Preloading runnable_avg_period should make the load_contrib a little
less sensitive to this behaviour.
> >> IMHO, I could only tell that for the new forked task, it could be run if current
> >> task already be set as need_resched, and preempt_schedule or
> >> preempt_schedule_irq
> >> is called.
> >> Since the set slice to avg behavior would not affect this task's vruntime,
> >> and hence cannot make current running task be need_sched, if
> >> previously it cannot.
> >> Could you help correct if I am wrong at somewhere? ....
> >> Thanks,
> > --
> > Thanks
> > Alex
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/