Estimating Electricity Usage

This page documents how the Ada Carbon Monitoring system estimates electricity usage (kWh) from CPU metrics.

Input Data

We have access to CPU time metrics from Prometheus:

  • busy_cpu_seconds - Total CPU-seconds spent in “busy” states
  • idle_cpu_seconds - Total CPU-seconds spent in “idle” states

These come from node_cpu_seconds_total metric aggregated by mode.

Power Model

Different CPU states consume different amounts of power:

State Power (per core) Description
Busy 12 W Active computation (user, system, nice, irq, softirq, steal)
Idle 1 W Waiting for work (idle, iowait)

Why These Values?

  • 12W busy: Conservative estimate based on typical server CPU power per core under load
  • 1W idle: Minimal power for core maintenance when not processing

These are configurable in ada-carbon-monitoring-api.ini:

[POWER]
busy_power_w = 12.0
idle_power_w = 1.0

Formula

Convert CPU seconds to energy (kWh):

Energy (kWh) = (busy_power × busy_seconds + idle_power × idle_seconds) / 3,600,000

Where:

  • Energy is in kilowatt-hours (kWh)
  • Power is in watts (W)
  • Time is in seconds (s)
  • 3,600,000 = 1000 (W→kW) × 3600 (s→h)

Implementation 

class ElectricityEstimator:
    """Estimates electricity usage from CPU seconds."""

    def __init__(
        self,
        busy_power_w: float = 12.0,
        idle_power_w: float = 1.0
    ):
        self.busy_power_w = busy_power_w
        self.idle_power_w = idle_power_w

    def estimate_busy_usage_kwh(self, busy_cpu_seconds: float) -> float:
        """Estimate energy for busy CPU time."""
        watt_seconds = self.busy_power_w * busy_cpu_seconds
        return watt_seconds / 3_600_000

    def estimate_idle_usage_kwh(self, idle_cpu_seconds: float) -> float:
        """Estimate energy for idle CPU time."""
        watt_seconds = self.idle_power_w * idle_cpu_seconds
        return watt_seconds / 3_600_000

    def estimate_total_kwh(
        self,
        busy_cpu_seconds: float,
        idle_cpu_seconds: float
    ) -> float:
        """Estimate total energy usage."""
        return (
            self.estimate_busy_usage_kwh(busy_cpu_seconds) +
            self.estimate_idle_usage_kwh(idle_cpu_seconds)
        )

Location: ada-carbon-monitoring-api/src/calculators/electricity_estimator.py

Example Calculations

Example 1: Typical Workspace (1 hour)

A workspace running for 1 hour with mixed usage:

Metric Value
Busy CPU seconds 1,800 s (30 minutes busy)
Idle CPU seconds 1,800 s (30 minutes idle) 
busy_kwh = 12 × 1,800 / 3,600,000 = 0.006 kWh
idle_kwh = 1 × 1,800 / 3,600,000 = 0.0005 kWh
total_kwh = 0.0065 kWh

Example 2: High-Utilization Job

An intensive computation running for 8 hours:

Metric Value
Busy CPU seconds 28,800 s (8 hours busy)
Idle CPU seconds 0 s 
busy_kwh = 12 × 28,800 / 3,600,000 = 0.096 kWh
idle_kwh = 0
total_kwh = 0.096 kWh

Example 3: Idle Workspace

A workspace sitting idle for 24 hours:

Metric Value
Busy CPU seconds 0 s
Idle CPU seconds 86,400 s (24 hours) 
busy_kwh = 0
idle_kwh = 1 × 86,400 / 3,600,000 = 0.024 kWh
total_kwh = 0.024 kWh

API Endpoint 

GET /carbon/electricity/estimate?busy_cpu_seconds=1800&idle_cpu_seconds=1800

Response:

{
  "busy_kwh": 0.006,
  "idle_kwh": 0.0005,
  "total_kwh": 0.0065,
  "busy_power_w": 12.0,
  "idle_power_w": 1.0
}

Comparison with TDP Method

An alternative approach uses CPU TDP (Thermal Design Power):

# TDP method (simpler, less accurate)
usage_kwh = cpu_seconds_total * cpu_tdp_w / 3_600_000
Method Pros Cons
Busy/Idle Accounts for idle time Requires mode breakdown
TDP Simple calculation Overestimates idle power

We use the busy/idle method for more accurate estimates.

Multi-Core Considerations

The calculation implicitly handles multiple cores:

  • Prometheus aggregates CPU seconds across all cores
  • 8 cores running for 1 hour = 28,800 CPU seconds
  • Power values are per-core, so the math works out

Validation

Compare calculated values against known benchmarks:

Scenario Expected Calculated Error
1 core, 1 hour, 100% busy 0.012 kWh 0.012 kWh 0%
4 cores, 1 hour, 50% busy 0.026 kWh 0.026 kWh 0%
8 cores, 8 hours, mixed ~0.4 kWh 0.38 kWh 5%

Next Steps

After calculating electricity usage, the next step is to calculate carbon footprint by multiplying by carbon intensity:

→ See Carbon Footprint