Abstract: To improve the performance of data parallel pipelines on computational grids, the algorithm of simulated annealing is used to optimize scheduling. An experimental method is used to determine the initial control parameter, and a random method is used to choose the next scheduling scheme. By adjusting the stability times of the best scheduling scheme, the trade-offs between optimization cost and optimization effect can be made. Experiments show the simulated annealing here can effectively optimize the scheduling of pipelines.

Key words: Grid, Scheduling, Data parallel pipeline, Simulated annealing

摘要： 为改进计算网格上数据并行流水线的性能，用模拟退火算法优化调度，用一个实验性方法确定初始控制参数，用一个随机方法选择下一个调度方案。调整最佳调度方案的稳定次数，可实现优化开销和优化效果之间的折中。实验显示，该文的模拟退火可有效优化流水线的调度。

关键词: 网格, 调度, 数据并行流水线, 模拟退火

CLC Number: 

• TP393