Skip to main content

Windowing

OpenDT aggregates tasks into time windows for batch simulation. This approach balances simulation accuracy with computational efficiency.

How Windows Work​

Tasks are assigned to windows based on their submission timestamp:

Window 0: [00:00 - 00:05)  →  Tasks with submission_time in this range
Window 1: [00:05 - 00:10)  →  ...
Window 2: [00:10 - 00:15)  →  ...

The window size is controlled by simulation_frequency_minutes in the configuration.

Window Lifecycle​

  1. Open: Window accepts tasks as they arrive
  2. Fill: Tasks accumulate in the window buffer
  3. Close: Heartbeat timestamp exceeds window end time
  4. Simulate: All accumulated tasks are processed by OpenDC
  5. Output: Results are written to agg_results.parquet

Heartbeats​

Heartbeat messages are synthetic timestamps published by dc-mock to signal time progression. They enable deterministic window closing even when no tasks arrive during a period.

Without heartbeats, windows would only close when new tasks arrive, leading to unpredictable timing in sparse workloads.

Cumulative Simulation​

OpenDT uses cumulative simulation: each window simulates all tasks from the beginning of the workload, not just tasks in that window.

This approach ensures accurate power predictions because:

  • Long-running tasks affect power across multiple windows
  • Scheduler state persists between windows
  • Host allocations remain consistent

Tuning Window Size​

Window SizeTrade-offs
Small (1-5 min)More granular results, higher computational cost
Large (30-60 min)Fewer simulations, less granular results

For long experiments, larger windows reduce total runtime while still capturing overall trends.