Task Reconfiguration Problems in High-Performance Distributed-Memory Machines

本研究主要探讨分布式记忆机器在发生故障时，如何重新配置任务模组的技术。问题包括：(1)局部重构和全局重构之间的权衡；(2)恢复成功后支持重构和任务计算所需的通信；(3)容错嵌入；(4)分析故障节点存在时网络的残余结构；(5)在Paralex 10维超立方体上开发的故障模式和算法性能的广泛模拟。

The proposed research concerns the exploration of techniques for reconfiguration of task modules on distributed-memory machines when faults occur. The problems include (1) trade offs between local and global reconfiguration; (2) communications required to support reconfiguration and task computation after successful recovery; (3) fault-tolerant embedding; (4) analysis of the residual structure of networks in the presence of faulty nodes, and (5) extensive simulations of fault patterns and the performance of algorithms developed on a Paralex 10-dimensional hypercube.