直连网络中的动态自适应容错路由算法  

Dynamic adaptive fault-tolerant routing methodology in direct networks

作　　者：董志凌[1] 朱永华[2] 池强[1] 

机构地区：[1]上海大学计算机工程与科学,上海200072 [2]上海大学计算中心,上海200436

出　　处：《计算机工程与设计》2010年第15期3367-3370,共4页Computer Engineering and Design

基　　金：国家863高技术研究发展计划基金项目(2005AA104031)

摘　　要：针对超级计算机的直连网络发生链路或节点故障,需要在保证一定网络性能的前提下在线完成网络重构的问题,提出一种直连网络中的动态自适应容错路由算法。该算法能够承受空间和时间上任何数目故障的组合,使发生故障的系统经过自动重构后能够维持较优的计算能力。该算法不牺牲任何健康节点,并且无需过多硬件资源。模拟结果表明,在网络无故障的情况下几乎不影响直连网络的性能,数据结果也表明了故障恢复的实时性和有效性。A dynamic adaptive fault-tolerant routing methodology for direct interconnection networks is presented to the condition that the direct network in the supercomputer occurs the fault in the links or nodes and they must be reconstructed online without loss of much performance.The methodology can tolerate any number of failures regardless of their spatial and temporal combinations.The remained system still behaves well after the automatic network reconfiguration finished online.The methodology doesn＇t need to sacrifice any health node and too much hardware resources.Performances of the scheme proposed are evaluated that the efficiency is almost as the same as the normal situation and data also confirmed the results of real-time recovery and effectiveness.

关 键 词：动态容错 自适应 路由 直连网络 超级计算机 

分 类 号：TP393.02[自动化与计算机技术—计算机应用技术]