Practical Schemes for Fat-Tree Network Construction

Acknowledgments Tom Knight originally suggested the idea of using fat-tree structures as a means for scaling Transit networks. Tom Knight and Henry Minsky were invaluable for discussion and criticism of these ideas throughout the development process. Discussions with Tom Leighton, Charles Leiserson, and Alex Ishii were valuable to my understanding of the more theoretical aspects of network organization. Pat Sobalvarro was helpful in providing the basic tools for the probabilistic analysis. Fred Drenckhahn is responsible for fleshing out many of the ideas in the Transit packaging scheme. Thanks to Tom Knight, Mike Bolotski, and Ellen Spertus for commenting on early versions of this paper. [8] Tom Leighton and Bruce Maggs. Expanders might be practical: fast algorithms for routing around faults on multibutterflies. – level of fat-tree – probability that a connection is routed through tree level – probability that a connection is routed to a particular processor through tree level