A multi-objective reinforcement learning algorithm for deadline constrained scientific workflow scheduling in clouds

作     者：Yao QIN Hua WANG Shanwen YI Xiaole LI Linbo ZHAI Yao QIN;Hua WANG;Shanwen YI;Xiaole LI;Linbo ZHAI

作者机构：School of Computer Science and TechnologyShandong UniversityJinan 250101China Shanghai Police CollegeShanghai 200137China School of SoftwareShandong UniversityJinan 250101China School of Information Science and EngineeringLinyi UniversityLinyi 276005China School of Information Science and EngineeringShandong Normal UniversityJinan 250014China 

出 版 物：《Frontiers of Computer Science》 (中国计算机科学前沿（英文版）)

年 卷 期：2021年第15卷第5期

页      面：25-36页

核心收录：

学科分类：081203[工学-计算机应用技术] 08[工学] 0835[工学-软件工程] 0812[工学-计算机科学与技术（可授工学、理学学位）] 

基　　金：the National Natural Science Foundation of China(Grant No.61672323) the Fundamental Research Funds of Shandong University(2017JC043) the Key Research and Development Program of Shandong Province(2017GGX10122,2017GGX10142,and 2019JZZY010134) the Natural Science Foundation of Shandong Province(ZR2019MF072) 

主　　题：workflow scheduling energy saving multiobjective reinforcement learning deadline constrained cloud computing 

摘      要：Recently,a growing number of scientific applications have been migrated into the *** deal with the problems brought by clouds,more and more researchers start to consider multiple optimization goals in workflow ***,the previous works ignore some details,which are challenging but *** existing multi-objective work-flow scheduling algorithms overlook weight selection,which may result in the quality degradation of ***,we find that the famous partial critical path(PCP)strategy,which has been widely used to meet the deadline constraint,can not accurately reflect the situation of each time ***-flow scheduling is an NP-hard problem,so self-optimizing algorithms are more suitable to solve *** this paper,the aim is to solve a workflow scheduling problem with a deadline *** design a deadline constrained scientific workflow scheduling algorithm based on multi-objective reinforcement learning(RL)called *** uses the Chebyshev scalarization function to scalarize its *** method is good at choosing weights for *** propose an improved version of the PCP strategy called *** sub-deadlines in MPCP regularly update during the scheduling phase,so they can accurately reflect the situation of each time *** optimization objectives in this paper include minimizing the execution cost and energy consumption within a given ***,we use four scientific workflows to compare DCMORL and several representa-tive scheduling *** results indicate that DCMORL outperforms the above *** far as we know,it is the first time to apply RL to a deadline constrained workflow scheduling problem.