simulator

Core simulation engine that consumes workload data and produces power predictions using OpenDC.

Purpose

The simulator aggregates tasks into time windows, invokes the OpenDC simulator, and outputs power consumption predictions. It maintains cumulative state to ensure accurate long-running simulations.

Processing Flow

1. Consume - Read tasks from dc.workload topic
2. Aggregate - Group tasks into time windows
3. Close - When heartbeat timestamp exceeds window end
4. Simulate - Invoke OpenDC with all tasks from beginning
5. Output - Write results to agg_results.parquet

Cumulative Simulation

Each window simulates all tasks from the beginning of the workload, not just tasks in that window. This ensures accurate power predictions for long-running workloads.

OpenDC Integration

The simulator invokes the OpenDC binary for power calculations:

opendc/bin/OpenDCExperimentRunner

Input Files

Created for each simulation run:

• experiment.json - OpenDC configuration
• topology.json - Datacenter topology
• tasks.parquet - Task definitions
• fragments.parquet - Task execution profiles

Output Files

Parsed after simulation:

• powerSource.parquet - Power consumption over time
• host.parquet - Host-level metrics

Output

Results are written to data/<RUN_ID>/simulator/:

simulator/
├── agg_results.parquet
└── opendc/
    └── run_<N>/

agg_results.parquet

Column	Description
timestamp	Simulation timestamp
power_draw	Predicted power in Watts
carbon_intensity	Grid carbon intensity

Configuration

Setting	Type	Default	Description
simulation_frequency_minutes	int	5	Window size for task aggregation

Larger windows mean fewer OpenDC invocations but less granular results.

Logs

make logs-simulator