Self-adaptation and mutual adaptation for distributedrnscheduling in benevolent clouds

摘要

Joint service involving several clouds is an emerging form of cloud computing. In hybrid clouds, thernschedulers within 1 cloud must not only self-adapt to the job arrival processes and the workloadrnbut also mutually adapt to the scheduling polices of other schedulers. However, as a combinatorialrnoptimization problem, scheduling is challenged by the adaptation to those dynamics andrnuncertain behaviors of the peers. This article studies the collaboration among benevolent cloudsrnthat are cooperative in nature and willing to accept jobs from other clouds. We take advantagernof machine learning and propose a distributed scheduling mechanism to learn the knowledge ofrnjobmodel, resource performance, and others’ policies. Without explicit modeling and prediction,rnmachine learning guides scheduling decisions based on experiences. To examine the performancernof our approach,we conducted simulation using the SP2 jobworkload log of the SanDiego SupercomputerrnCenter under a test bed based on agent-based systems—SWARM. The results validaternthat our approach has much shorter mean response time than 5 typical dynamic schedulingrnalgorithms—opportunistic load balancing, minimum execution time, minimum completion time,rnswitching algorithm, and k-percent best. A better collaboration in hybrid cloud is achieved byrnfull adaptation.