标题：Toward better service performance management via workload prediction
作者：Moussa, Hachem ;Yen, I-Ling ;Bastani, Farokh ;Dong, Yulin ;He, Wei
作者机构：[Moussa, Hachem ;Yen, I-Ling ;Bastani, Farokh ] University of Texas at Dallas, Dallas; TX; 75080, United States;[Dong, Yulin ;He, Wei ] Shandong Unive 更多
会议名称：16th International Conference on Services Computing, SCC 2019, held as Part of the Services Conference Federation, SCF 2019
会议日期：25 June 2019 through 30 June 2019
来源：Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
关键词：Performance management; Service composition; Service execution; Service performance; Workload prediction
摘要：In this paper, we consider managing service performance starting from the composition time, aiming to reduce the risk of execution failures during service composition. We use ARIMA to predict workloads of the services at the time when they are likely to be invoked and subsequently predict the response time and chances that the requests for accessing the services may be declined due to admission control. The in-depth analysis can help avoid timing failures during service execution. However, these analyses may incur overhead and we introduce a two-phase composition algorithm to reduce the potential overhead. Our system also considers continuous monitoring and service recomposition to greatly increase the probability of completing the service execution within the deadline. Experimental results show that our service management approach can greatly improve the success rate for meeting the deadline. © Springer Nature Switzerland AG 2019.