CSR: Small: Self-Monitoring Virtual Machines for Performance Guarantees in Public Clouds

凭借其庞大的计算资源池和多路复用，云为大型组织和小型企业提供了更低的信息技术成本、更轻的管理负担和快速扩展资源的前景。但是，公共云中的多租户使得关键计算资源（如处理器、内存、I/O设备和存储）在由不同用户操作的虚拟机之间共享。因此，在公共云中运行的应用程序的性能可能会因为共享计算资源的争用而受到相邻虚拟机的影响。尽管如此，现有的云性能监控工具并不提供对硬件资源的可见性；云用户别无选择，只能盲目地在这些公共服务上运行计算，希望性能不会受到负面影响，例如，邻居耗尽资源的应用程序。缺乏性能保证是所有云用户要充分利用云计算的经济效益所面临的一个障碍，特别是那些应用程序需要运行时环境的稳定性和可预测性的用户。这个提议的项目旨在通过开发新的技术来解决这个问题，这些技术允许云用户在没有云提供商帮助的情况下监视物理云服务器上的资源争用。具体来说，建议的工作需要设计、实现和评估自我监视虚拟机，它利用嵌套虚拟化技术和侧信道分析技术来监视共享计算资源中的争用，并主动迁移嵌套虚拟机以避免严重的性能下降。该项目将在以下方面产生更广泛的社会影响：首先，通过拟议的项目将开发新的教育工具。具体来说，预期研究的结果之一将是亚马逊机器图像，可以在公共云之上创建派生云。通过这种衍生云，操作系统或系统安全课程的学生可以获得云计算的实践经验；他们还可以访问嵌套的虚拟化环境来进行操作系统内核开发。其次，该项目将被纳入NSF的LSAMP（路易斯·斯托克斯少数民族参与联盟）计划，以帮助增加代表性不足的少数民族学生的招募、保留和获得STEM学位，并提高代表性不足的少数民族学生在系统研究中的参与度。第三，该项目将生产能够在公共云中自我监控虚拟机的开源工具，这些工具将以源代码（可在项目主页上获得）和Amazon机器映像的形式提供，以鼓励对已开发技术的适应和采用。

With its massive pooling and multiplexing of computing resources, the cloud offers both large organizations and small businesses the prospect of lower information technology costs, lighter administrative burdens, and rapid scaling of resources. However, multi-tenancy in public clouds makes critical computing resources, such as processor, memory, I/O devices, and storage, shared among virtual machines that are operated by different users. The performance of applications running in public clouds, therefore, may be affected by their neighboring virtual machines due to contention on the shared computing resources. Nonetheless, existing cloud performance monitoring tools do not offer visibility into hardware resources; cloud users have no choice but to blindly run computations on these public services in the hope that the performance is not negatively affected, for instance, by the neighbor's resource-depleting applications. The lack of performance guarantees is a hurdle faced by all cloud users to fully embrace the economic benefit of cloud computing, and especially by those whose applications demand stability and predictability of the runtime environments. This proposed project aims to solve this problem by developing novel techniques that allow cloud users to monitor the resource contention on the physical cloud servers without the help of the cloud providers. Specifically, the proposed work entails the design, implementation, and evaluation of self-monitoring virtual machines, which leverage the nested virtualization technology and side-channel analysis techniques to monitor the contention in shared computing resources and proactively migrate nested virtual machines to avoid severe performance degradation.The project will make broader societal impacts in the following aspects: First, new education tools will be developed through the proposed project. Specifically, one of the outcomes of the intended research will be Amazon machine images with which a derivative cloud can be created on top of public clouds. Enabled by this derivative cloud, students of the operating systems or system security courses can obtain hands-on experience with cloud computing; they will also have access to nested virtualization environments to conduct operating system kernel development. Second, the project will be integrated into NSF's LSAMP (Louis Stokes Alliances for Minority Participation) program, to help increase underrepresented minority student recruitment, retention, and attainment of STEM degrees, and also to enhance the participation of underrepresented minority students in system research. Third, the project will produce open-source tools that enable self-monitoring VMs in public clouds, which will be made available in the form of source code (available on the project homepage) and Amazon machine images, to encourage adaptation and adoption of the developed techniques.