Hierarchical networks with Byzantine Robustness

In traditional networks, a single malicious (“Byzantine”) packet switch can cause global disruption; for instance, by giving incorrect routing information, flooding the network with traffic, or forwarding data incorrectly. Previous work (which we'll call NPBR for “Network Protocols with Byzantine Robustness”) presented a network design resilient to Byzantine failures, that guaranteed that nodes A and B can communicate, with some fair share of bandwidth, provided that at least one honest path connects them. NPBR, for reasons described in this paper, only works in a fairly small, flat network. This paper presents a network design that not only provides the same guarantees as NPBR, in a large hierarchical network, but provides additional guarantees, which we will describe in the paper. Furthermore, our design does not require any router to keep state larger than necessary for its portion of the network hierarchy. In this paper we summarize NPBR, explain why it does not extend to a hierarchy, and then present a design suitable for a hierarchy.