Byzantine robustness, hierarchy, and scalability

In [20] (which we'll refer to as NPBR, for “Network Protocols with Byzantine Robustness”) the term “Byzantine Robustness” was defined as the property that guarantees that any two parties, S and D, can converse with each other, with a fair share of bandwidth, even if some of the trusted components in the network (e.g., routers, links) are behaving maliciously, provided that at least one correctly functioning path connects S and D. NPBR only works in a fairly small, flat network. In order to scale to larger sizes, networks need to be organized in hierarchies. This paper presents a network design (which we will refer to as “HBR” for Hierarchical Byzantine Robustness) that provides the same correctness guarantee as NPBR, but our design works in a hierarchical network. Furthermore, our design does not require any node to keep state larger than necessary for its portion of the network hierarchy. Also, our design extends to protection of the destination from being overwhelmed with too many simultaneous connections, with the network queuing connection requests so that each eventually completes. In this paper we summarize NPBR, explain why it does not extend to a hierarchy, and then present our design.