Abstract: In response to the problems of poor resource utilization and inconvenient services caused by the uneven distribution of cloud service resources among regions,a collaborative scheduling model of service resources based on cloud alliances was proposed,which had considered the interests of cloud service providers and customers as the supply and demand sides of service resources.The impact of service demand scheduling in cloud federation on the two was quantified and analyzed based on modeling the two using queueing theory with cloud service provider profit and customer wait time as the guide.The idea of general customer demand and emergency customer demand was further introduced in the construction of the service demand scheduling strategy,and the differences between the two types of demand were modeled and analyzed separately.Deep Q-learning algorithm was used to optimally solve and simulate the service resource collaborative scheduling model.The simulation results validated the effectiveness of the proposed method,which protected the interests of both cloud service providers and customers while promoting the circulation and sharing of service resources.

Key words: cloud federation, deep reinforcement learning, queueing theory, demand scheduling

摘要： 针对云服务资源区域间分布不均衡的现状而导致的资源利用率不高、服务不便等问题,结合考虑云服务商和顾客作为服务资源供需双方的利益诉求,提出一种基于云联盟的服务协同调度模型。首先,以云服务商利润和顾客等待时间为导向,使用排队论对二者进行建模,在此基础上量化分析了云联盟中的服务需求调度对二者造成的影响。其次,在服务需求调度策略的构建中进一步引入普通顾客需求和紧急顾客需求的思想,并针对两类需求的差异性分别进行了建模和分析。最后,利用深度强化学习(Deep Q-learning)算法对服务资源协同调度模型进行优化求解和模拟仿真,仿真结果验证了所提方法的有效性,在推动服务资源的流通和共享的基础上同时保障了云服务商和顾客的利益诉求。

关键词: 云联盟, 深度强化学习, 排队论, 需求调度