Tailoring Genetic Algorithm for Resource Scheduling in Many-Core Processors
Xiande Hu, Jingming Li, Jiaxing Cheng · Advances in computer science research · 2015
With the development of multi-core and many-core processors, how to leverage the hardware improvement to boost application performance remains a challenge for ensuring the continuously improvement of computing and data processing capabilities.Specifically, in this paper, we focus on how to efficiently assign many-core resources to computing tasks to improve the processing capability of systems.We propose to employ Genetic Algorithm (GA) for many-core resource scheduling.Considering the issues with GA, such as the low convergence speed, low efficiency of the search process, and limitations of the genetic operators, we propose two modifications.First, to increase the discrimination of fitness value and therefore accelerate the convergence speed, we introduce the idea from Simulated Annealing (SA) to design the fitness function.Second, to deal with the limitations of genetic operators, we introduce Cellular Automata (CA), and therefore to overcome the problem of premature convergence.With the extensive experiments, our modified algorithm exhibits fast convergence and efficient performance for many-core scheduling.