非一致性存储结构架构下服务器物理资源竞争问题

Research on resource contention problem on non uniform memory access server

为了提升云服务器的负载应对能力,虚拟化技术被进一步应用于云数据中心的资源调度和资源管理。然而,虚拟化技术在提升服务器资源利用率的同时,也面临着用户类型多样化以及多任务间不同资源类型需求混合而带来的挑战,服务器的物理资源竞争问题日益严峻。该文以非一致性存储结构(NUMA)服务器为对象,分别研究了一般性物理资源竞争及其多Socket之间的资源竞争问题。提出了一般性资源竞争的度量模型,并给出了包括多Socket资源竞争在内的应对策略。实验表明,该模型能够有效表征服务器物理资源的竞争程度,相应的策略算法能够降低物理资源争夺而带来的性能干扰,提升任务运行效率。

In order to improve the ability of handling workload,the virtualization technology has been further applied in resource scheduling and resource management.Although virtualization technology has improved the ability and performance of cloud servers,it also confronts various user types and combination of applications with different resource demands,thus the problem of resource contention has become severer nowadays.In the scope of non uniform memory access(NUMA)server,this paper focuses on the resource contention issue both on general physical resources and the specific resource between multiple sockets.This paper proposes a general model to quantify the degree of resource contention.Then an appropriate strategy is introduced consequently to cope with the resource contention problem including the contention between multiple sockets.Experiments in this paper show that this model can characterize the resource contention effectively.Meanwhile the corresponding strategy can also reduce performance interference significantly thus leads to higher efficiency of running applications.