In order to construct a metric which approximates power consumption, we need to understand how accurately we can predict power consumption based on CPU usage.

We should build an approximation using pstates, and one without pstates, and quantify how accurate these approximations are.
 

In order to construct a metric which approximates power consumption, we need to understand how accurately we can predict power consumption using various approaches.

We should build and quantify the accuracy of the following approaches:
• Task Durations
  • With pstate information
  • Without pstate information
  • Roughly predicting pstate information
• Sampling Tasks every N ms.
  • With pstate information
  • Without pstate information
  • Roughly predicting pstate information

We intend to use telemetry to quantitatively evaluate these approaches.
 

In order to construct a metric which approximates power consumption, we need to understand how accurately we can predict power consumption using various approaches.

We should build and quantify the accuracy of the following approaches:
• CPU Clock Task Durations
  • With pstate information
  • Without pstate information
  • Roughly predicting pstate information
• Wall Clock Task Durations
  • With pstate information
  • Without pstate information
  • Roughly predicting pstate information
• Sampling Tasks every N ms.
  • With pstate information
  • Without pstate information
  • Roughly predicting pstate information

We intend to use telemetry to quantitatively evaluate these approaches.
 

In order to construct a metric which approximates power consumption, we need to understand how accurately we can predict power consumption using various approaches.

We should build and quantify the accuracy of the following approaches:
• CPU Clock Task Durations
  • With pstate information
  • Without pstate information
  • Roughly predicting pstate information
• Wall Clock Task Durations
  • With pstate information
  • Without pstate information
  • Roughly predicting pstate information
• Sampling Tasks every N CPU ms.
  • With pstate information
  • Without pstate information
  • Roughly predicting pstate information
• Sampling Tasks every N ms.
  • With pstate information
  • Without pstate information
  • Roughly predicting pstate information

We intend to use telemetry to quantitatively evaluate these approaches.