RRecherchily
← Back to projects
ActiveApril 2026

Energy-Efficient Scheduling in Heterogeneous Computing Clusters via Multi-Objective Meta-Heuristic Optimization

Data centers and heterogeneous computing clusters (CPU, GPU, FPGA) consume considerable energy with often suboptimal scheduling policies. This project formulates scheduling as a multi-objective optimization problem (minimizing energy, makespan, and SLA violations) and proposes hybrid meta-heuristic algorithms (NSGA-III + adaptive local search) outperforming classical heuristics.

SchedulingMeta-heuristicsHPCEnergy efficiencyNSGA-III

Team

NR
Nadia RAHMANI
Researcher
KT
Karim TOUATI
Researcher

Methodology

  1. Model a realistic heterogeneous cluster (32 CPU nodes + 8 GPUs + 4 FPGAs) with measured energy consumption profiles and workloads from Google Cluster traces.
  2. Implement NSGA-III with problem-specific crossover and mutation operators, coupled with adaptive local search (VNS — Variable Neighborhood Search).
  3. Compare with six reference algorithms: Round-Robin, Min-Min, HEFT, MOHEFT, SPEA2, and MOEA/D across 200 varied workload instances.
  4. Sensitivity analysis on parameters (population size, mutation rate) and scalability study up to 1,000 simulated nodes.

Evaluation metrics

  • Hypervolume of the Pareto solution set — improvement of > 12% over MOHEFT.
  • Average energy consumption reduction compared to Round-Robin — target > 25%.
  • SLA violation rate (deadline miss rate) — target < 3%.

Budget

$300 in compute credits for optimization campaigns, $200 for access to real workload traces.

Expected deliverables

  • Open-source heterogeneous cluster simulator with standardized benchmark interface.
  • Python library of multi-objective meta-heuristic algorithms for scheduling.
  • Paper submitted to NeurIPS 2027 Workshop on Optimization.
← Back to projects