Power Models

The CPUPowerModel defines how CPU utilization translates to power consumption. Different model types suit different hardware characteristics.

Model Types

Model	Description	Best For
mse	Mean Squared Error based model	General purpose, default
asymptotic	Non-linear curve with asymptotic behavior	High-utilization workloads
linear	Linear interpolation between idle and max power	Simple approximations

Parameters

All power models share these common parameters:

Parameter	Type	Description
idlePower	float	Power draw at 0% utilization (Watts)
maxPower	float	Power draw at 100% utilization (Watts)

MSE Model

The MSE model uses a calibration factor to scale predictions:

Parameter	Description
calibrationFactor	Scaling multiplier for power calculation

Asymptotic Model

The asymptotic model provides non-linear scaling:

Parameter	Description
asymUtil	Curve coefficient (0-1), controls the shape of the utilization curve

Example Configuration

{
  "cpuPowerModel": {
    "modelType": "mse",
    "idlePower": 25.0,
    "maxPower": 174.0,
    "calibrationFactor": 10.0
  }
}

Calibration

When calibration is enabled, OpenDT automatically tunes power model parameters to minimize the error between predicted and actual power consumption.

The calibrator runs parallel simulations with different parameter values, calculates MAPE (Mean Absolute Percentage Error) against actual measurements, and publishes the best-performing configuration.