服务质量约束下的虚拟集群资源分配策略研究

摘要

To effectively allocate computer resources under virtual cluster-based high performance computing environments to satisfy users' different QoS requirements, this paper presents a novel dynamically-adjustable resources allocate strategy. The strategy regards the slowdown degree of users' job as a new metric of QoS and the key role of resources allocation. From the perspective of resource providers, this strategy guarantees low expenditure of users and high resource profit of users service. To implement the strategy, a mixed bin packing algorithm is also presen-. ted, which is the combination of the traditional bin-packing algorithm and some general algorithms. The experimental results show that the mixed bin packing algorithm is more effective than the previous algorithms, especially in a large application scale background.%为了有效实现基于虚拟机技术的高性能计算环境资源调度与分配,以满足用户不同服务质量请求,提出了一种动态可调节的资源分配策略.该策略以用户任务被放缓程度为核心指标,并允许用户根据该指标提出相应的服务质量约束,从资源提供者的角度保证了用户获得相关服务的开销最低,同时总的资源收益最大.对该分配模式下的静态资源分配问题进行了形式化的描述,并结合该策略提出混合背包算法.实验结果表明:面对大规模应用时,该算法较其它一般性算法更易得到效能较优的资源分配方案.