论文部分内容阅读
在当今信息时代，随着移动设备、互联网应用以及云计算模式的快速发展，数据中心已成为社会基础设施。然而数据中心面临资源利用率与应用服务质量之间的矛盾，一方面通过多个应用同时在数据中心部署实现资源共享能有效提高资源利用率，另一方面多个应用共享资源又会出现相互干扰，严重影响应用的服务质量。因此，目前企业不得不采用预留额外资源以保障延迟敏感的关键应用服务质量，这导致数据中心的利用率很低。并且，随着多核技术的发展，单个服务器内的资源越来越多，其上混合部署的应用数目也在不断增加，更加剧这种矛盾。如何解决资源利用率与应用服务质量之间的矛盾，是数据中心面临的核心挑战之一，同时也为计算机系统结构研究带来很多机遇。文章主要介绍了数据中心所面临的上述矛盾以及一些研究进展，最后介绍了资源可编程体系结构 PARD(Programmable Architecture of Resourcing on-Demand)思想，从硬件上支持资源容量隔离与性能隔离，从而保障多应用混合环境下关键应用的服务质量，允许更大程度混合部署应用以提高数据中心资源利用率。“,”Datacenters have become the fundamental infrastructure of modern internet services such as searching, social networking and online shopping. The resource utilization of typical datacenters, however, is quite low. The primary reason of such low utilization is that datacenter operators have to upload resources to online services that are latency sensitive. Co-locating multiple applications in a shared datacenter is an effective approach to improve resources utilization but can result in degradation of application’s quality of service (QoS) because of interference between online services and other background workloads. Therefore, QoS guarantee method in a shared datacenter becomes the key technology for improving resourcing utilization of datacenters. In this paper, a survey of research efforts on this topic was ifrstly presented and then a programmable architecture for resourcing on-demand (PARD) was reviewed. The goal of PARD is to make hardware to support resources management so that the QoS of key applications can be guaranteed while multiple applications execute together in shared datacenters.