摘要
超立方是分布存储系中最常用的结构.在以往的工作中,人们已经提出了不少容错寻径算法.然而,还没有考虑Hn中|F|2n-2的情形.在一个含有故障结点集F的n维超立方网络Hn中,|F|4n-24,(s0,d0),(s1,d1)是其中任意两对非故障结点,如果,(1)对于v∈V(Hn),有|A(v,Hn-F)|6.(2)沿着某一维k(0kn-1),可将Hn分割成两部分:(d0∈)Hn-1,0和(d1∈)Hn-1,1,且|F∩Hn-1,i|2n-12(i=0,1),则一定存在两条互不相交的路径P(si,di),使得|P(si,di)|H(si,di)+12(i=0,1).并且,这两条路径可以并行地求得.我们给出了相应的容错寻径算法,其时间复杂性为t=O(n·|F|).
Hypercube is one of the most commonly used topological structures for distributed memory systems. There have been a number of fault tolerant routing algorithm proposed in previous work. While, none of these previous schemes can work when there are more than 2 n -2 processors failed in a system of multiprocessors connected with n dimensional hypercube network. In this paper, this problem is deeply studied.Let H n be a n dimensional Hypercube network with fault node set F,|F|4n-24,(n>10),(s 0,d 0) and ( s 1,d 1 ) be two pairs of nodes in H n-F . If they satisfy, (1) |A(v, H n-F)|6 for arbitrary v∈V(H n) .(2) H n can be parted into two subgraphs, ( d 0∈)H n-1,0 and (d 1∈)H n-1,1 , and |F∩H n-1,i |2n-12, for i=0,1. Then,there are two disjoint fault free paths P(s i,d i) ,such that, |P(s i,d i)|H(s i,d i)+12,(i=0,1), and the two paths can be parallel constructed.The corresponding fault tolerant routing algorithm which its time complexity is t=O(n· |F|) is given in this paper.
出处
《计算机学报》
EI
CSCD
北大核心
1999年第2期120-125,共6页
Chinese Journal of Computers
基金
北京建工学院青年科研基金
关键词
图论
互连网络
寻径算法
超立方网络
Graph theory, interconnection networks, fault tolerance, routing algorithm, hypercube.