27th Annual IEEE Conference on Local Computer Networks, 2002. Proceedings. LCN 2002.
Download PDF

Abstract

System area networks have been developed to address the needs of computing clusters. Myricom?s Myrinet architecture is one of the predominant technologies in this area. One of the key issues for SANs is fault-tolerant routing. Myrinet provides Mapper software to discover and maintain network topology. Myrinet?s Mapper is centralized, susceptible to probe packet deadlock, and does not incorporate host monitoring. We propose an alternative mapper called AM3. AM3 is hierarchical, reduces the number of probe packets, and incorporates host monitoring. We have implemented a prototype version of AM3 on Chiba City, Argonne National Lab?s 512 CPU linux cluster. Keywords: Network Fault Management, System Area Network, Myrinet, Active Mapper
Like what you’re reading?
Already a member?
Get this article FREE with a new membership!