Scheduling for heterogeneous Systems using constrained critical paths

详细信息	查看全文 | 推荐本文 |

• 作者：Minhaj Ahmad Khan ^{minhajahmad@gmail.com}
• 关键词：Scheduling algorithms ; Heterogeneous scheduling ; List scheduling ; Static scheduling ; Parallel applications
• 刊名：Parallel Computing
• 出版年：2012
• 期刊代码：99_01678191
• 类别：cp
• 出版时间：April-May, 2012
• 卷：38
• 期：4-5
• 页码：175-193
• 文件大小：1047 K

摘要

A complex computing problem may be efficiently solved on a system with multiple processing elements by dividing its implementation code into several tasks or modules that execute in parallel. The modules may then be assigned to and scheduled on the processing elements so that the total execution time is minimum. Finding an optimal schedule for parallel programs is a non-trivial task and is considered to be NP-complete.

For heterogeneous systems having processors with different characteristics, most of the scheduling algorithms use greedy approach to assign processors to the modules. This paper suggests a novel approach called constrained earliest finish time (CEFT) to provide better schedules for heterogeneous systems using the concept of the constrained critical paths (CCPs). In contrast to other approaches used for heterogeneous systems, the CEFT strategy takes into account a broader view of the input task graph. Furthermore, the statically generated CCPs may be efficiently scheduled in comparison with other approaches.

The experimentation results show that the CEFT scheduling strategy outperforms the well-known HEFT, DLS and LMT strategies by producing shorter schedules for a diverse collection of task graphs.