Honours projectsSample project

GridSense

Schedules computer-lab workloads around load-shedding forecasts so long-running jobs finish before the power goes off.

Category
Energy
Year
2024
Student
Pieter van Wyk
Supervisor
Prof Daniel Kriel

Research question

Simulations and model training in the labs are lost when load-shedding starts mid-run, wasting hours of compute and student time.

Approach

Published load-shedding schedules feed a constraint solver that places queued jobs into windows where they can complete.

Key features

  • Reads the published load-shedding schedule automatically
  • Estimates job length from previous runs
  • Suggests the earliest safe start time

Evaluation

Simulated on a semester of lab usage, the scheduler was compared with first-come, first-served queuing.

GridSense
Open original
GridSense project poster