Latency Aware Task Allocation in Heterogeneous Multi-Core System using Whale Optimization Algorithm

Abstract

Heterogeneous multi-core systems are increasingly used in latency-sensitive computing environments because they combine different processing units such as high-performance CPUs, low-power cores, GPUs, NPUs, and DSPs. However, the diversity of these processing units makes task allocation difficult, especially when execution time, communication delay, processor availability, energy consumption, and workload dependencies must be considered together. This study develops a latency-aware task allocation framework for heterogeneous multi-core systems using the Whale Optimization Algorithm (WOA). The objective is to formulate task scheduling as a discrete task-to-core mapping problem and evaluate the effectiveness of WOA under independent, DAG-based, real-time, and energy-aware workload conditions. The proposed method adapts the continuous WOA into a discrete integer-encoded scheduling model, where each whale represents a complete task-to-processor assignment. A latency-based fitness function was used for WOA-Latency, while an extended weighted objective was applied for WOA-EnergyAware to examine latency-energy trade-offs. The experimental evaluation compared the proposed WOA variants with Random, Round-Robin, MET, MCT, Min-Min, Max-Min, HEFT, PEFT, GA, PSO, GWO, ACO, and a WOA-MCT hybrid scheduler. The evaluation included repeated simulation runs, statistical reporting, convergence analysis, scalability testing, deadline miss analysis, and parameter sensitivity assessment. The results show that WOA-Latency outperforms several naïve and metaheuristic baselines in some workload settings, particularly Random, Round-Robin, GA, PSO, and GWO. However, deterministic heuristics such as MCT and Min-Min performed better for independent-task workloads, while HEFT and PEFT achieved stronger results for DAG-based scheduling. The energy-aware WOA variant demonstrated the framework’s potential for multi-objective optimisation, although with some latency trade-off. Overall, the study concludes that WOA is a flexible and extensible optimisation framework for heterogeneous task allocation, especially in complex, nonlinear, energy-aware, and multi-objective scheduling scenarios.