[Paper Review] Multiprocessor Scheduling Using Parallel Genetic Algorithm

Nourah Al-Angari, Abdullatif Alabdullatif|arXiv (Cornell University)|Sep 24, 2012

Distributed and Parallel Computing Systems6 references3 citations

TL;DR

This paper proposes a synchronous master-slave parallel genetic algorithm (GA) to accelerate multiprocessor scheduling, addressing the high computational cost of fitness evaluation in sequential GAs. By distributing fitness evaluations across multiple processors, the approach achieves significant speedup on complex, high-generation scheduling problems, outperforming sequential execution in both runtime and scalability.

ABSTRACT

Tasks scheduling is the most challenging problem in the parallel computing. Hence, the inappropriate scheduling will reduce or even abort the utilization of the true potential of the parallelization. Genetic algorithm (GA) has been successfully applied to solve the scheduling problem. The fitness evaluation is the most time consuming GA operation for the CPU time, which affect the GA performance. The proposed synchronous master-slave algorithm outperforms the sequential algorithm in case of complex and high number of generations problem.

Motivation & Objective

• To address the performance bottleneck in genetic algorithm-based multiprocessor scheduling caused by time-consuming fitness evaluation.
• To improve the efficiency and scalability of scheduling in parallel computing environments.
• To design and implement a synchronous master-slave parallel GA architecture for distributed fitness evaluation.
• To evaluate the performance gain of the parallel approach over sequential GA on large-scale scheduling problems.
• To demonstrate the feasibility and effectiveness of parallelization in accelerating NP-hard scheduling optimization.

Proposed method

• A synchronous master-slave architecture is employed, where the master coordinates the GA process and distributes individuals to slave processors.
• Each slave processor independently evaluates the fitness of assigned individuals using a predefined scheduling cost function.
• Fitness evaluation is the primary computational workload, and its parallelization is the core optimization target.
• The algorithm proceeds in synchronized generations, with the master collecting results and updating the population.
• The approach is implemented and tested on a cluster environment to evaluate performance on complex scheduling instances.
• The method leverages message passing for inter-process communication, ensuring coordination and load balancing.

Experimental results

Research questions

• RQ1 Can parallelization of fitness evaluation in a genetic algorithm significantly reduce runtime for multiprocessor scheduling?
• RQ2 How does the performance of the parallel master-slave GA compare to a sequential GA on large-scale scheduling problems?
• RQ3 Does the synchronous execution model maintain solution quality while improving speed?
• RQ4 What is the scalability of the proposed parallel GA with increasing numbers of processors and problem complexity?
• RQ5 Can the parallel approach handle high-generation, complex scheduling problems that are infeasible for sequential execution?

Key findings

• The proposed parallel genetic algorithm achieves faster convergence and reduced execution time compared to the sequential implementation.
• The synchronous master-slave model effectively balances load and minimizes idle time across processors.
• Performance improvement is most significant on problems with high computational cost per fitness evaluation and large population sizes.
• The algorithm demonstrates scalability, with consistent speedup as the number of processors increases.
• The solution quality remains competitive with sequential GA, indicating that parallelization does not compromise optimization effectiveness.
• The approach is particularly effective for complex scheduling problems requiring many generations, where fitness evaluation dominates runtime.