SOFTWARE:"Cluster-based Runtime Support for Data-Intensive Online Applications"

数据密集型应用程序（如数据挖掘）需要大量的计算和I/O资源。随着万维网的普及和集群技术的进步，将这些数据密集型应用程序实时地提供给大量用户已经成为一个可以实现的目标。在线应用程序（如Internet服务）与脱机应用程序的不同之处在于对高度并发工作负载的性能敏感性和可用性需求。在高度并发的工作负载下，为这些应用程序实现高性能、可靠性和高效的资源管理是一项挑战。该项目解决了在为高度并发的网络服务和在线应用程序构建高性能和可靠的运行时支持时的系统和编程级问题。这个项目有三个主要的研究重点。首先，研究基于集群的系统架构，重点关注整体系统可伸缩性/可用性以及负载平衡和数据复制支持。其次，该项目研究了编程和运行时支持，以简化涉及集群上并发和并行数据操作的在线应用程序的开发和部署。第三，以服务质量和整体资源利用效率为目标，构建基于集群的资源管理框架。在整个项目中，将努力检查数据密集型应用并验证所提议工作的有效性。随着通过系统开发和应用评估获得更多经验，将对其他问题进行调查。开发的软件将公开发布，以供其他应用程序的潜在使用。由于越来越多的数据密集型服务通过内部网和互联网提供，预计开发的结果将产生广泛的影响。本提案中描述的研究将直接有利于pi机构的本科和研究生教育的教学改进，并增强学生在基于集群的计算方面的经验。

Data-intensive applications such as data mining demand significant compute and I/O resources. With the ubiquity of World Wide Web and the advances in clustering technologies, it becomes a reachable goal to bring these data-intensive applications to a large number of users in real-time. Online applications such as Internet services differ from their offline counterpart in the performance sensitivity to highly concurrent workload and the availability requirement. It is challenging to achieve high performance, reliability, and efficient resource management for these applications under highly concurrent workload.This project addresses the system and programming-level issues in buildinghigh performance and reliable runtime support for highly concurrentnetwork services and online applications. There are three main researchthrusts in this project. First, cluster-based system architectures are studied with a focus on overall system scalability/availability as well as load balancing and data replication support. Secondly, this project investigates programming and runtime support to simplify the development and deployment of online applications that involve concurrent and parallel data manipulation on clusters. Thirdly, a cluster-based resource management framework will be developed with the combined goal of quality-of-service and overall resourceutilization efficiency. Throughout this project, efforts will be made toexamine data-intensive applications and validate the effectiveness of theproposed work. Additional issues will be investigated as more experienceare gained through system development and application evaluation.The developed software will be released in public for potential use of otherapplications.Since more and more data-intensive services are being madeavailable through intranets and the Internet, it is expected that thedeveloped results will have broad impacts. Research described in this proposal will directly benefit instruction improvement for undergraduate and graduate education at PIs' institutions and enhance students' experience in cluster-basedcomputing.