CAREER: High-Performance Algorithms for Scientific Applications

并行计算长期以来提供了非常高的性能，但它只在很小的应用范围内实现了这一点。在大型分布式内存系统层面上开发并行性受到消息传递成本的阻碍，而共享内存系统的规模仍然很小。然而，随着对称多处理器(SMPS)的出现，最大规模的共享内存正在成为一种可用的商品。高性能千兆网络允许可扩展的应用程序在SMP的大型集群上运行。在未来五到十年内，SMP集群将可能成为可扩展高性能计算的主要体系结构；然而，到目前为止，在这些SMP集群上支持有效的并行计算的工作还很少。我们进行的初步工作表明，有可能改进现有的SMP集群编程方法。在这个职业奖项中，目标是为SMP集群开发、实施、评估和优化算法，用于不规则(例如，基于字符串、树和图形)的计算，这些算法将在SMP集群的典型配置上提供显著的加速，并随处理器数量的增加而优雅地扩展。这项研究将研究支持不规则计算的新算法和基本例程库，主要是基于树和图的计算，以及如何利用PRAM算法中的理论研究的新见解。基因组学、生物信息学和计算生态学中的科学驱动问题将成为这项研究的重点。该项目的教育部分包括指导高中生和少数族裔学生，通过演讲和论文展示研究成果，以及在关键会议和研讨会上展示教程。之前的辅导已经产生了几个在地方和国家比赛中获得第一名的个人和团体团队，并且辅导将继续与少数群体合作，如全国黑人工程师协会和美国普韦布洛原住民学生团体。

Parallel computing has long offered the promise of very high performance, but it has deliveredonly in a narrow range of applications. Exploiting parallelism at the level of large distributed-memorysystems is hampered by the cost of message-passing, while shared-memory systems remain mostlysmall-scale. With the advent of symmetric multiprocessors (SMPs), however, shared-memory on amodest scale is becoming an available commodity. High-performance gigabit networks allow scalable applications to run on large clusters of SMPs. Over the next five to ten years, clusters of SMPs will likely be the predominant architecture for scalable high-performance computing; however, little work has been done to date to support effective parallel computing on these SMP clusters.Preliminary work we have conducted indicates that it is possible to improve upon current programmingmethods for SMP clusters. In this career award the goal is to develop, implement, assess, andrefine algorithms for SMP clusters for irregular (e.g., string-, tree-, and graph-based) computations thatwill deliver significant speedups on typical configurations of SMP clusters and scale gracefully withthe number of processors. The research will investigate new algorithms and a library of basic routines tosupport irregular computations, mostly tree- and graph- based, along with new insights on how to leveragethe theoretical research in PRAM algorithms. Science-driven problems in genomics,bioinformatics, and computational ecology will provide the focus for this research.The education component of this project includes mentoring high school and minority students, diseminating research results through talks and papers, and presenting tutorials at key conferences andworkshops. Prior mentoring has produced several individual and group teams that have won first place in both local and national competitions and mentoring will continue activities with minority groups, such as the National Society of Black Engineers and Native American Pueblo student groups.