Create consumption profile based on punctual measure

[da-ekchajzer]

Context

Today, usage impacts are measured by multiplying a duration, an impact factor, and an electrical consumption : impact = electrical_consumption*duration*impact_factor

We focus in this issue on how to identify electrical_consumption.
There are 2 ways today in the API :

1. Given by the user, in case the user has a sensor like boagent, scaphandre or a wattmeter. We should keep this behavior.

2. Using a Workload object which links a percentage of workload to an electrical consumption. See doc for more info.

If the user have the electrical consumption he can use this type of query :

"max_power": 510,
"workload": {
      "100": {
        "time": 0.15,
        "power": 1.0
      },
      "50": {
        "time": 0.55,
        "power": 0.7235
      },
      "10": {
        "time": 0.2,
        "power": 0.5118
      },
      "idle": {
        "time": 0.1,
        "power": 0.3941
      }
    }

This is kind of dump since the user could directly provide the electrical consumption following the first method.

If the user doesn't have his electrical consumption (which is the normal case), he can provide a workload object without power. The power will be smart completed. We use this behavior in the case of AWS instances with Teads punctual consumption profile.

"workload": {
      "100": {
        "time": 0.15,
      },
      "50": {
        "time": 0.55,
      },
      "10": {
        "time": 0.2,
      },
      "idle": {
        "time": 0.1,
      }
    }

Limits

With such behavior :

1. We can only use a specific load repartition : [idle, 10, 50, 100] (In some cases, we only have min and max value, we could have more precise repartition in the case of stress test for instance)
2. We cannot give a specific workload (for example My server have been running at 34% for 2 hours)
3. The workload object is hard to understand and manipulate

Solution

Instead of having punctual workload objects we should use functions generated from the regression of punctual measures.

The idea is to use a functions : workload_to_electrical_conso which returns an electrical consumption from a given workload :
workload_to_electrical_conso(workload(%)) = electrical_consumption

TODO

Since this problem is complex it will be divided into several issues

• Provide a way to generate function based on punctual consumption profile #86
• Smart-complete electrical_consumption with consumption_profil(), single workload data #87
• Smart-complete electrical_consumption with consumption_profil(), multiple workload #88
• Smart-complete consumption_profile based on CPU model #89

[da-ekchajzer]

Implemented in #113