NGS: Wisconsin DOVE: Distributed Optimizing Virtual Environment

rastislav bodikwisconsin大学：Wisconsin DOVE：分布式优化虚拟环境Wisconsin DOVE提出提高现代分布式和并行应用程序的性能，以及提高应用程序开发人员和系统维护人员的生产力。基于分布式控制、动态程序优化和硬件支持的性能监控方面的最新进展，我们的项目将构建一个分布式优化虚拟环境（DOVE），其功能将源于两个主要创新。VM-in-OS范例。Dove将在操作系统中植入一个优化的虚拟机（VM），它能够分析正在运行的程序，并将分析与基于硬件的性能监控相关联，从而实现垂直集成，跨越应用程序、内核和硬件。支持vm的操作系统将密切关注上面的应用程序以及下面的硬件和网络，因此它将能够更智能、更自适应地调度资源。一群优化的虚拟机。预计未来的分布式应用程序将由用Java（或类似的移动语言）编写的分布式组件组装而成，我们建议将单个分布式组件底层的vm组织成一个集群，其中vm交换有关其客户端的概要分析和程序分析信息。通过支持集群中的分布式vm之间的“八卦”，我们将能够计算分布式应用程序的运行时通信和依赖配置文件，并相应地执行应用程序的动态重新分区。

EIA-0103670Rastislav BodikUniversity of WisconsinNGS: Wisconsin DOVE: Distributed Optimizing Virtual EnvironmentWisconsin DOVE proposes to improve the performance of modern distributed and parallel applications as well as increase the productivity of application developers and system maintainers. Building on recent advances in distributed control, dynamic program optimization, and hardware-supported performance monitoring, our project will build a Distributed Optimizing Virtual Environment (DOVE) whose power will stem from two primary innovations.The VM-in-OS paradigm. Dove will implant into the operating system an optimizing virtual machine (VM), whose ability to analyze a running program and correlate the analysis with hardware-based performance monitoring will achieve vertical integration, spanning the application, the kernel, and the hardware. The VM-enabled operating system will be intimately aware of both the application above and the hardware and network below, and hence it will be able to schedule resources more intelligently and adaptively.A clan of optimizing virtual machines. Expecting that future distributed applications will be assembled from distributed components written in Java (or a similar mobile language), we propose to organize the VMs underlying the individual distributed components into a clan, in which the VMs exchange profiling and program-analysis information about their clients. By supporting "gossip" among the distributed VMs in a clan, we will be able to compute a run-time communication and dependence profile of the distributed application and, in response, perform a dynamic repartitioning of the application.