同构多核/众核处理器任务分配自适应模拟退火算法

Adaptive Simulated Annealing Algorithm for Task Assignment on Homogeneous Multi/Many-core Processors

随着多核/众核处理器核心数快速增加,任务分配解空间急剧增大,降低近似解的相对偏差越来越难。提出一种自适应模拟退火算法,建立了模拟退火算法中参数与优化环境任务数和核心数的关系。核心数的增加不但可以有效降低近似解的相对偏差,而且使任务分配算法具有较高的环境自适应能力。与较近研究成果相比较,在16核心时,自适应模拟退火算法迭代次数增加41%,相对偏差降低86%。

With rapid increasing of the number of cores in multi/many-core processors,the task assignment solution space increases sharply so that reducing the relative deviation of the approximate solution becomes more and more difficult.An adaptive simulated annealing algorithm was put forward by establishing the relationship of the algorithm parameters and the number of optimized environment tasks and cores.The increasing of core number can not only effectively reduce the relative deviation of the approximate solution,but also shows high adaptability for the environment.Experiments reveal that on the 16 cores platform,the adaptive simulated annealing algorithm iterations are increased by 41％,but the relative deviation is decreased by 86 ％ versus the recent research results.