[dpdk-dev] [PATCH v4 1/2] lib/librte_power: traffic pattern aware power control

[Liang, Ma]

Hi Kevin,
     Many thanks for your feedback.
     Please check my comments below. 

On 13 Sep 10:46, Kevin Traynor wrote:
> 
> Thanks for following up. It's allowing it to run without a training
> phase which is what I thought could be problematic from an application
> view, so that's nice. I'm not sure if it's much less effective without
> that training phase etc, but the comment was focused on having a forced
> training phase, so that is resolved now as it is not required.
> 
I have re-worked the patch, therefore,  the simple app will run without training as default option
> I'm still not sure I see the use cases for the options where there *is*
> a training type phase but it's difficult to know and considering it's
> experimental, if you feel there are some potential use cases and
> justification to add it, then fine with me.
> 
However, the Training  still is necessary. 
The mechanism need 2 anchor point. 
1.  The max empty poll the system can reach without any real traffic. 
    That's is Maximum capability we use as a base line. 
2.  When the empty poll number drop to zero, that indicate the system 
    is 100% busy due to always get work to do. 
3.  When the empty poll number drop to certain point(e.g. 30% of Max)
    the mechanism will move to next frequency.

Without the Training phase, it's very hard to use normal traffic payload 
to get the absolute anchor point due to the traffic type, payload size, 
processor micro-arch, cache size etc, too many moving parts. 

I use default value which I think will be OK for mainstream xeon. 
if user use a very different system(e.g. arm), they still need re-run training phase.

the training can be triggered by a parameter option. 

> I have a few comments on the API, which I'll reply directly to the patch.
> 
> thanks,
> Kevin.
> 
> > Regards,
> > Dave.
> > 
> > 
> > 
> > 
> > 
> > 
> > 
> > 
> > 
> > 
>