Techniques for Fault-Tolerant Communication in Parallel Computers

并行计算机中的容错通信技术

• 批准号: 9525887
• 负责人: Robert Cypher
• 金额: $ 15.38万
• 依托单位: Johns Hopkins University
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 1996
• 资助国家: 美国
• 起止时间: 1996-01-01 至 1999-12-31
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=9525887&HistoricalAwards=false
• 关键词: Techniques; Fault; Tolerant; Communication; Parallel

The project studies techniques for performing efficient communication in parallel computers that have faulty components. Parallel computers with a large number of identical processors have a great potential for fault-tolerance. The main challenge that must be overcome in order to exploit this potential for fault- tolerance is the fact that faults also affect communications between processors. Hardware faults in processors, routers, and communication links force messages to take new paths that avoid the faulty components. In addition, the mapping of computations from faulty processors to nonfaulty ones changes the pattern of communication. As a result, faults can have a major impact on the performance, and even correctness, of the communication subsystem in a parallel computer. Specifically, the project focuses on the following two key issues; creating efficient, deadlock-free routing algorithms for parallel computers with faulty components, and developing efficient mappings of communication-intensive real-time applications to parallel computers that contain faults. This research will result in more efficient use of parallel computers and a better understanding of how their inherent potential for fault-tolerance can be utilized.

该项目研究在具有故障组件的并行计算机中执行有效通信的技术。 具有大量相同处理器的并行计算机具有很大的容错潜力。 为了利用这种容错潜力，必须克服的主要挑战是故障也会影响处理器之间的通信。 处理器、路由器和通信链路中的硬件故障会迫使消息采取新的路径，以避开故障组件。 此外，从故障处理器到非故障处理器的计算映射改变了通信模式。 因此，故障可能对并行计算机中通信子系统的性能甚至正确性产生重大影响。 具体而言，该项目侧重于以下两个关键问题：创建高效的，无死锁的路由算法的并行计算机故障组件，并开发通信密集型实时应用程序的高效映射到包含故障的并行计算机。 这项研究将导致更有效地利用并行计算机和更好地了解如何利用其固有的容错潜力。