CSR: Small: Big Memory: Exploring Memory Management Mechanisms and Policies for Rack-Scale Computers

大内存应用程序（包括数据分析、内存对象缓存和内存数据库）是数据中心资源的主要消耗者。 此外，这些应用程序可以有效地使用更多的内存比一台计算机可以提供。 扩展这些应用程序的传统方法涉及多台计算机之间的并行性。 但是，这可能无法有效地使用资源，因为处理、网络和内存的比例是由底层计算机的配置决定的，而不是应用程序。 这个项目探讨了分散式服务器，或“机架规模的计算机”。这些服务器由机架内的处理器，内存和I/O设备组成，可以根据其驻留应用程序的需求灵活地分配给虚拟机。这个项目侧重于分散式服务器的一个方面：管理机架内的大型内存池。 具体来说，使用此池可以动态地向虚拟机添加内存和从虚拟机删除内存。这将使应用程序能够处理突发负载，而无需启动/关闭整个虚拟机。 为了跨虚拟机有效地管理内存，虚拟机管理程序需要更好地理解客户机操作系统和应用程序如何使用内存。 因此，我们正在探索虚拟机管理程序、客户操作系统和应用程序内部和之间的接口、机制和策略，这些将支持使用大型分散内存。这项研究是重新思考计算机是什么以及如何构建一个更好地服务于数据科学和分析应用程序的一步，这些应用程序可以从大量内存中受益。

Large-memory applications, including data analytics, in-memory objectcaches, and in-memory databases, are a major consumer of data centerresources. Moreover, these applications can effectively use morememory than a single computer can provide. Traditional approaches toscaling these applications involve parallelism across multiplecomputers. However, that may not use resources cost-effectively, asthe ratio of processing, networking, and memory is determined by theconfiguration of the underlying computers, not the application. Thisproject explores disaggregated servers, or "rack-scale computers".These servers consist of pools of processors, memory, and I/O deviceswithin a rack that can be flexibly allocated to virtual machines basedon the needs of their resident applications.This project focuses on one aspect of disaggregated servers: managinga large pool of memory within the rack. Specifically, using this poolto dynamically add memory to and remove memory from virtual machines.This will enable applications to handle bursty load without the needto spin up/down entire virtual machines. In order to effectivelymanage memory across virtual machines, the hypervisor will need abetter understanding of how memory is being used, by both the guestoperating systems and applications. Therefore, we are exploring theinterfaces, mechanisms, and policies within and between thehypervisor, guest operating system, and application that will enablethe use of large disaggregated memories. This research is one step towards rethinking what a computer is and how to construct one that better serves data science and analytics applications that can benefit from large amounts of memory.