Re: [PATCH v2 0/2] Adds cpu power accounting per-pid basis.

On Thu, May 14, 2015 at 11:34 PM, Heinrich Schuchardt
[off-list ref] wrote:

On 15.05.2015 02:12, Ruchi Kandoi wrote:

quoted

These patches add a mechanism which will accurately caculate the CPU power
used by all the processes in the system. In order to account for the power
used by all the processes a data field "cpu_power" has been added in the
task_struct.

Hello Ruchi,

could you, please, explain why the CPU power consumption per task
information is needed. Please, consider that the CPU causes only part of
the total system power consumption which also comprises GPU, cooling,
RAM, etc.

In order to accurately account for the battery used by each of the
process, keeping a track of how long the process ran is not
sufficient. Since running at different frequency has varying power
consumption, we want to track a power number which takes into
consideration the frequency as well as the core on which it was
running. There are similar efforts for other subsystems too to account
for the power used by each process which can then accurately be
aggregated for an application.

The patch series increases the memory size of the kernel, the memory
consumption per thread and the thread switching time. So, please,
introduce a configuration switch to enable/disable the function.

Yes, configuration can be added. Will update that in the next patch.

quoted

This field adds power for both the system as well as user
time. cpu_power contains the total amount of charge(in uAmsec units) used
by the process.

Is there any reasonable way to assign the power consumption to a single
task if multiple tasks are executed on the same core at the same time
(e.g. using hyperthreading)?

I think the power will be accounted for both the processes on their
respective cores. With hyperthreading, as far as kernel is concerned
they are running on different cores and the time for all the tasks
will be accounted appropriately and hence power. Correct me if I am
wrong.

quoted

This model takes into account the frequency at which the
process was running(i.e higher power for processes running at higher
frequencies). It requires the cpufreq_stats module to be initialized with
the current numbers for each of the CPU core at each frequency. This will
be initialized during init time.

This does not account for power consumption depending on anything else
but frequency, e.g. floating point commands consuming more power than NOPs.

Currently we have been able to get power numbers for a core when they
are active and running at a particular frequency. Agreed that will be
a better and more accurate mode.But getting the power numbers for the
type of instruction and keeping track of number of such instructions
will be cumbersome.

Best regards

Heinrich Schuchardt

quoted

Ruchi Kandoi (2):
  cpufreq_stats: Adds sysfs file
    /sys/devices/system/cpu/cpufreq/current_in_state
  sched: cpufreq: Adds a field cpu_power in the task_struct

 drivers/cpufreq/cpufreq_stats.c | 191 +++++++++++++++++++++++++++++++++++++++-
 include/linux/cpufreq.h         |   8 ++
 include/linux/sched.h           |   2 +
 kernel/fork.c                   |   1 +
 kernel/sched/cputime.c          |   7 ++
 5 files changed, 207 insertions(+), 2 deletions(-)