Computing on Clusters: Data Abstractions, Runtime Systems, and Resource Management

涉及视觉、动画和多媒体协作的真实交互式多媒体很可能成为未来计算机应用的一个重要方面。 这种应用程序固有的可扩展并行性加上其计算需求使其成为SMP和SMP集群的理想候选者。 这些应用程序是高度动态的，并且具有新的要求，例如需要对应用程序产生的数据进行时间索引。 在并行编程常用的抽象和这样的需求之间存在不匹配，导致应用程序开发人员在正确性和性能方面都付出了相当大的努力。 建议的研究的目标是设计和评估并行编程的抽象，运行时机制，和资源管理算法，适合这种复杂的应用程序的需求。 具体而言，预计将收到以下捐款：（a）设计和评估高级并行编程抽象，以及评估在使用这种抽象时性能和可编程性之间的权衡，（B）设计和评估运行时机制（诸如时间管理和分布式垃圾收集），以便在集群中有效地实现这样的抽象，以及（c）设计和评价集群中注重应用的资源管理机制。

Umakishore RamachandranCCR-9972216Realistic interactive multimedia involving vision, animation, and multimedia collaboration is likely to become an important aspect of future computer applications. The scalable parallelism inherent in such applications coupled with their computational demands make them ideal candidates for SMPs and clusters of SMPs. These applications are highly dynamic and have novel requirements such as the need for temporally indexing the data produced by the application. There is a mismatch between the abstractions commonly available for parallel programming and such requirements, leading to a considerable effort on the application developer both in terms of correctness and performance. The goal of the proposed research is the design and evaluation of parallel programming abstractions, runtime mechanisms, and resource management algorithms suited to the needs of such complex applications. In particular, the following contributions are expected: (a) Design and evaluation of high-level parallel programming abstractions, and the evaluation of the tradeoff between performance and programmability in using such abstractions, (b) Design and evaluation of runtime mechanisms (such as time management and distributed garbage collections) for implementing such abstractions efficiently in a cluster, and (c) Design and evaluation of application-conscious resource management mechanisms in a cluster.