CSR-PDOS: Optimizing the Client/Server Environment Subject to User Satisfaction

该项目通过让系统软件直接优化以满足个人用户的满意这一概念，重新审视交互式客户机/服务器计算中的系统问题。在系统软件级别对特定配置或参数选择的满意程度因用户而异。该项目测试了这样一个假设，即一个薄的、非侵入式的用户界面，在某些情况下与学习算法相结合，可以将单个用户的满意度水平或其他指导直接传达给系统软件，并提供足够的细节，以允许系统软件有效地为单个用户定制其决策。为了收集关于该假设的证据，该项目正在将该概念应用于在瘦客户机、桌面替换系统和现代web应用程序等上下文中出现的几个系统问题。其中包括：电源管理中的动态电压和频率缩放算法；对远程显示系统的猜测；线程、进程和虚拟机调度；以及多核处理器中的线程分配/迁移。这些领域的工作是统一的：(1)系统软件的有效用户界面的设计、实现和评估（通过用户研究），(2)寻求确定显式和隐式用户反馈的适当组合，以及(3)确定学习的作用，以尽量减少显式反馈的数量。如果这个假设成立，它将阐明一种构建更能让个人用户满意的系统的新方法。

This project reexamines systems problems in interactive client/server computing through the concept of having systems software optimize directly for the expressed satisfaction of the individual user. Satisfaction with a particular configuration or choice of parameters at the system software level varies dramatically from user to user. The project tests the hypothesis that a thin, nonintrusive user interface, in some cases combined with learning algorithms, can convey the individual user's satisfaction level, or other guidance, directly to the systems software with sufficient detail to allow the systems software to effectively customize its decisions to that individual user. To gather evidence about the hypothesis, the project is applying the concept to several systems problems that arise in contexts such as thin clients, desktop replacement systems, and modern web applications. These include: dynamic voltage and frequency scaling algorithms in power management; speculation in remote display systems; thread, process, and VM scheduling; and thread assignment/migration in multicore processors. The work in these areas is unified through: (1) the design, implementation, and evaluation (through user studies) of effective user interfaces to systems software, (2) seeking to determine an appropriate mix of explicit and implicit user feedback, and (3) determining the role of learning to minimize the amount of explicit feedback. If the hypothesis holds, it will illustrate a new way to build systems that are more satisfying for individual users.