Re: [PATCH] docs: add example for psi info

Re: [PATCH] docs: add example for psi info

[Jonathan Corbet]

Xavier Claude <contact@xavierclaude.be> writes:

> Provide an non zero example with explanation of the values.
>
> Signed-off-by: Xavier Claude <contact@xavierclaude.be>
> ---
>  Documentation/accounting/psi.rst | 14 ++++++++++++++
>  1 file changed, 14 insertions(+)

This may be a good change, but you really want the memory-management
developers to look at it...adding linux-mm in the hopes that might make
it happen.

Thanks,

jon

> diff --git a/Documentation/accounting/psi.rst b/Documentation/accounting/psi.rst
> index d455db3e5808..dadab65f872d 100644
> --- a/Documentation/accounting/psi.rst
> +++ b/Documentation/accounting/psi.rst
> @@ -64,6 +64,20 @@ as well as medium and long term trends. The total absolute stall time
>  spikes which wouldn't necessarily make a dent in the time averages,
>  or to average trends over custom time frames.
>  
> +For example, the following output::
> +
> +	some avg10=7.52 avg60=12.59 avg300=24.82 total=2128219913967
> +	full avg10=5.70 avg60=10.56 avg300=22.40 total=2014527537356
> +
> +This means that for the last 10 seconds, 5.7% of the time (so 0.5
> +seconds) was lost on cpu cycles because all the running processes where
> +waiting on the resource. On the same period, some tasks where not
> +schedulable during 7.5% of the time (so 750ms) because they were waiting
> +on the resource. But for a longer period of 300 seconds, we see that
> +more than 20% of the time (so 60 seconds) is wasted due to resource
> +contention. Moreover, since the total time wasted because tasks were
> +waiting on the resource is 2014527.5 seconds.
> +
>  Monitoring for pressure thresholds
>  ==================================
>  
> -- 
> 2.55.0