CSR: Small: IRON: Reducing Workload Interference on Massively Parallel Platforms

基于蜻蜓和胖树设计的互连网络在高性能计算设施和数据中心中越来越占主导地位。管理此类网络的一个关键挑战是工作负载干扰。在多用户计算环境中，共享网络资源的应用程序之间的干扰可能导致事件（工作负载干扰、低生产率、自私的用户行为和糟糕的调度）相互加剧的恶性循环。这个项目旨在解决工作量干扰的恶性循环。该项目旨在开发一个软件框架来系统地分析和减轻蜻蜓和脂肪树系统的工作负载干扰。该项目的完成将创建新的干扰感知调度策略和可扩展的软件工具，用于蜻蜓和脂肪树网络的干扰分析和减少。从模拟和实验中收集的结果数据和工具将在开源许可下提供给广泛的社区。综合教育和推广计划将加强计算机科学课程，扩大代表性不足群体的参与，并扩大到以非洲裔美国人和拉丁裔为主的周边社区。提出的研究重点是三个相互关联的研究重点：(1)高保真网络模拟，以深入了解应用程序之间的网络干扰；(2)干扰感知调度，以避免或减轻应用程序之间的网络争用；(3)现实世界实验，定量测量生产系统上代表性应用程序的工作负载干扰，并评估干扰感知调度。由此产生的IRON将为推进大规模并行平台上工作负载干扰的研究提供一个集成的干扰分析和减少框架。项目网站（http://www.cs.iit.edu/~lan/iron.html）提供了有关项目的所有信息（出版物、软件工具、数据等）。该项目存储库将在项目完成后至少维护五年。

Interconnection networks based on dragonfly and fat tree designs are becoming increasingly dominant in high-performance computing facilities and data centers. A key challenge of managing such networks is workload interference. In a multi-user computing environment, interference among applications for shared network resources can cause a vicious cycle of events (workload interference, low productivity, selfish user behavior, and poor scheduling) aggravating each other. This project seeks to tackle the vicious circle of workload interference.This project aims to develop a software framework to systematically analyze and mitigate workload interference on dragonfly and fat tree systems. Completion of the project will create novel interference-aware scheduling policies and scalable software tools for interference analysis and reduction on dragonfly and fat tree networks. The resulting data and tools collected from simulations and experiments will be made available to the broad community under an open source license. An integrated education and outreach plan will enhance the Computer Science curriculum, broaden the participation by underrepresented groups, and outreach to the surrounding communities that are predominantly African-American and Latino.The proposed research focuses on three inter-related research thrusts: (1) high-fidelity network simulation to gain insights into network interference among applications, (2) interference aware scheduling to avoid or mitigate network contention among applications, and (3) real-world experiments to quantitatively measure workload interference of representative applications on production systems and assess interference aware scheduling. The resulting IRON will provide an integrated interference analysis and reduction framework for advancing the research of workload interference on massively parallel platforms. The project website (http://www.cs.iit.edu/~lan/iron.html) provides all the information about the project (publications, software tools, data, etc.). This project repository will be maintained for at least five years after the project completion.