On 29-Aug 10:06, Pavan Kondeti wrote:

Hi Patrick,

On Fri, Aug 25, 2017 at 3:50 PM, Patrick Bellasi
[off-list ref] wrote:

quoted

The util_avg signal computed by PELT is too variable for some use-cases.
For example, a big task waking up after a long sleep period will have its
utilization almost completely decayed. This introduces some latency before
schedutil will be able to pick the best frequency to run a task.

The same issue can affect task placement. Indeed, since the task
utilization is already decayed at wakeup, when the task is enqueued in a
CPU, this can results in a CPU running a big task as being temporarily
represented as being almost empty. This leads to a race condition where
other tasks can be potentially allocated on a CPU which just started to run
a big task which slept for a relatively long period.

Moreover, the utilization of a task is, by PELT definition, a continuously
changing metrics. This contributes in making almost instantly outdated some
decisions based on the value of the PELT's utilization.

For all these reasons, a more stable signal could probably do a better job
of representing the expected/estimated utilization of a SE/RQ. Such a
signal can be easily created on top of PELT by still using it as an
estimator which produces values to be aggregated once meaningful events
happens.

This patch adds a simple implementation of util_est, a new signal built on
top of PELT's util_avg where:

    util_est(se) = max(se::util_avg, f(se::util_avg@dequeue_times))

I don't see any wrapper function in this patch that implements this
signal. You want to use this signal in the task placement path as a
replacement of task_util(), right?

You right, I should update this changelog which is a bit misleading.

What I'm writing above is the way we combine a task's estimated
utilization with its util_avg. That's what you find in the code of the
following patches, but strictly speacking we do not have a wrapper
function.

Thanks,
Pavan

Cheers Patrick

-- 
#include <best/regards.h>

Patrick Bellasi