|
概要（英語）
|
Collective communication involves global data movement and global control among a group of processors in a parallel/distributed computer system. Among all collective communications, all-to-all personalized communication is one of the most dense communication patterns. In this paper, for n-dimensional faulty hypercubes, we propose an all-to-all personalized communication algorithm obtained by extending an all-to-all broadcasting algorithm fault-tolerant in the presence of up to \floor(n/2) faulty nodes. The proposed algorithm assumes a multiport communication and an availability of global fault information. The total time steps required to the algorithm is given as about 5 2^{n-1} in the presence of any multiple fault up to \floor(n/2), that is does not depend on the number of presented faults.
|