CAREER: Coordination Mechanisms for Performance-aware Virtualization in Clusters

虚拟化技术正在现代集群平台中迅速采用，这些平台支持对性能敏感的应用程序，如Web服务、实时金融交易、数据挖掘、电子科学和电子商务。用于虚拟机（VM）调度、分配和迁移的最先进技术通常将VM隔离对待，而不是在其他VM和它们不断交互的集群环境的上下文中对待。本项目研究集群范围的协调机制，利用虚拟机间和虚拟机集群的相互作用，以提高应用程序的性能和集群的利用率。该研究针对四个具体方向：（1）提高时间敏感型应用程序跨VM间交互的及时性，（2）虚拟化群集范围的内存资源以支持VM内的大内存和大数据集应用程序，（3）利用VM间的依赖性来指导VM迁移决策，以及（4）通过跨VM和虚拟群集的协调网络访问来减少网络结构争用。该项目的更广泛影响将是开发虚拟化技术，以更好地支持新兴数据中心和高性能计算集群中的性能敏感型应用程序。教育影响将体现在培训虚拟化技术专家、加强宾厄姆顿大学的实验计算机科学课程以及指导本科生、少数民族和高中生。

Virtualization technology is rapidly being adopted in modern cluster platforms that support performance-sensitive applications such as web services, real-time financial transactions, data mining, eScience, and E-commerce. State-of-the-art techniques for virtual machine (VM) scheduling, allocation, and migration often treat the VMs in isolation rather than within the context of other VMs and the cluster environment with which they constantly interact. This project investigates cluster-wide coordination mechanisms that exploit inter-VM and VM-cluster interactions to improve both application performance and cluster utilization. The research targets four specific directions: (1) improving timeliness across inter-VM interactions for time-sensitive applications, (2) virtualizing cluster-wide memory resources to support large memory and large dataset applications within VMs, (3) exploiting inter-VM dependencies to guide VM migration decisions, and (4) reducing network fabric contention through coordinated network access across VMs and virtual clusters. Broader impact of the project will be in the development of virtualization technologies that better support performance-sensitive applications in emerging data centers and high-performance computing clusters. Educational impact will be in training technology experts in virtualization, strengthening the experimental computer science curriculum at Binghamton University, and mentoring of undergraduate, minority, and high school students.