Scheduling Support for High Performance Clusters

集群正在成为高性能并行计算的主要和经济高效的平台。这些系统不仅被传统上需要超级计算机的科学应用所接受，而且还被部署以满足数据库、网络服务、多媒体和图形/可视化等领域的应用需求。传统的并行调度技术，如简单的空间共享或协同调度，它们针对吞吐量进行了更多的优化，可能无法处理这些新兴应用程序的响应需求。同时，完全让每个节点上的本机操作系统调度器做出独立的决策可能会对通信应用程序的性能产生不利影响。考虑到这些不同的问题，本提案将探索有效管理和协调集群节点间CPU资源的技术。本研究将开发一个框架来设计集群调度机制，将通信与工作集和I/O信息一起考虑在内，适应应用程序的约束和特性，并动态适应系统负载。将使用仿真、分析模型和实际实现，对这些机制进行实际工作负载评估。这项研究的结果将朝着使集群更加无处不在、多程序化、易于管理和用户友好的方向迈出重要的一步。

Clusters are becoming a predominant and cost-effective platformfor high performance parallel computing. These systems are gaining acceptancenot just for scientific applications, which have traditionally neededsupercomputers, but are also being deployed to meet the needs ofapplications in domains such as databases, web service, multimedia, and graphics/visualization. Traditional parallel scheduling techniquessuch as simple space sharing or coscheduling, which are optimized morefor throughput, may not be able to handle the responsiveness needs ofthese emerging applications. At the same time, leaving it entirelyto the native operating system scheduler at each node to make independentdecisions can have a detrimental effect on the performance of communicatingapplications. Considering these different issues, this proposal willexplore techniques for efficient management and coordination of the CPU resources across the nodes of a cluster. This research will developa framework to design cluster scheduling mechanisms taking communication into account together with working sets and I/O information, accomodatethe constraints and idiosyncrasies of an application, and adapt dynamicallyto the system load. The mechanisms will be evaluated with real workloads,using simulation, analytical models and actual implementations.The outcome of this research will take an important step towards makingclusters more ubiquitous, multiprogrammed, manageable and user-friendly.