Multicast is an important communication primitive for parallel programming. In this paper, the authors first analyze three kinds of reliable multicast algorithms generally used on Ethernet, then propose a new reliable algorithm called MAK(multicast acknowledgment) algorithm to obtain low communication latency. Measurements on it also show that the implementation achieves high performance.
1 Huang Y, McKinley P K. Efficient collective operations with ATM network interface support. In: Proceedings of the International Conference'96 on Parallel Processing. Los Alamitos, California: IEEE Computer Society Press. Aug. 1996. I-34~I-43
2 Anderson Thomas E, Culler David E et al. A case for NOW (networks of workstations). IEEE Micro, Feb. 1995,15(1):54~64
3 Lin Xiao-la, Ni Lionel M. Multicast communication in multicomputer networks. Technical Report, Michigan State University, 1989
4 Susanne E Hambrusch, Ashfaq A Khokhar, Liu Yi. Scalable S-to-P broadcasting on message-passing MPPs. In: Proceedings of the International Conference'96 on Parallel Processing. Los Alamitos, California: IEEE Computer Society Press. Aug. 1996. I-69~I-76
5 周明天,汪文勇.TCP/IP网络原理与技术.北京:清华大学出版社,1993
(Zhou Ming-tian, Wang Wen-yong. Network principle and technology with TCP/IP. Beijing: Tsinghua University Press, 1993)