虚拟计算系统中基于负载特征反馈的计算资源分配策略研究

项目名称： 虚拟计算系统中基于负载特征反馈的计算资源分配策略研究

项目编号： No.60873023

项目类型： 面上项目

立项/批准年度： 2009

项目学科： 金属学与金属工艺

项目作者： 万健

作者单位： 杭州电子科技大学

项目金额： 32万元

中文摘要： 虚拟化系统中，由于多个虚拟机共享相同的硬件资源，在高并发情况下，硬件资源往往在多个客户虚拟机之间竞争。因此，必须根据负载变化为虚拟机动态分配资源，满足多个虚拟机的资源需求的时间性能及服务公平性。但是，由于虚拟机环境下的硬件资源具有动态异构性，且虚拟机运行在虚拟机监控器之上，导致虚拟机监控器在进行资源调度时所获取的信息远少于传统的操作系统所掌握的进程信息，增加了负载特性识别、预测和推断的难度，并降低了其准确性。本课题从典型负载特征的角度，通过对运行时负载的进程信息的收集、发掘并建立典型负载随机模型，使资源管理器能够在运行时识别出不同负载类型特征和负载水平，应用经济学中市场模型和博弈模型，根据负载特征和负载水平动态调节计算资源的分配，应用不同的服务策略，自适应的协调控制资源分配，以保证每一个虚拟机只能得到预定的服务资源，保证服务公平性的同时避免出现服务瓶颈，同时对具有高优先权的虚拟机提供高性能服务。在Xen虚拟化系统上的实验结果表明，本课题提出的资源分配与管理算法，实现了虚拟计算系统的计算资源动态分配和高效利用，提高了虚拟计算系统的性能和服务质量，能够适用于大规模虚拟化系统。

中文关键词： 虚拟化；资源分配；负载特征；反馈；服务质量

英文摘要： In virtualized data centers and cloud computing systems, allocation and scheduling of virtual and physical resources raise more challenges and may have negative impacts on system performance due to the isolation, independence, heterogeneity, and conflicting operations among multiple virtual machines which share the same physical machine. Because virtual machines are on top of the Virtual Machine Monitor, this nature of virtualization results in contentions and conflicting under heavily loaded or consolidated situations. In order to accommodate as many as computing and service capabilities while still delivering performance guarantees, resource utilization and application performance should be trade-offed such that fewer resources will be utilized for a specific performance requirement. To achieve the goals of tradeoff between various performance metrics including responsiveness, Quality of Service (QoS), in this project we proposed adaptive resource allocation mechanism and strategies based on economic model and gaming theory. We use historical and real time performance profiling and estimation from agents within individual virtual machines for workload characterization. This approach does not require a highly accurate performance model in a virtualized system where workloads usually change frequently with time in some intervals. We also proposed a stochastic performance model of virtualized resources to provide performance guarantees and service differentiation in contending conditions through service differentiation. In these algorithms and prototypes the resources are allocated according to the demand as well as the performance of the targeted virtual machines based on the Sufferage aggregation and performance feedback. Experiments on a Xen based virtualized environment are conducted and evaluated. The results show that the proposed approach utilizes less CPU resources to achieve the same performance goals compared to the fixed over-provisioning of resources. The results also show that the feedback based algorithm is valid, effective and scalable for implementation in real virtualized environments.

英文关键词： Virtualization; Resource Allocation; Workload Characterization; Feedback; Quality of Service