Much of the Internet's critical DNS infrastructure -- several of the root servers and many top-level domains -- are being distributed with a technique called anycast. There are significant variations in network topology between different anycast systems.
There have been several studies of stability and query distribution of anycast networks. Most have focused on a single anycast system and may have told us more about the specific system studied than about anycast in general. A recent study by Liu, Hufffaker, Fomenkov, Brownlee, and claffy applied a consistent methodology to three different anycast systems and found significant differences in query distribution among the three.
This paper is an examination of the Liu et al. study from a network engineering perspective. It considers how network topology affects performance and why their research showed the results it did. It then looks at what we can infer from those results about the behavior of other anycast systems, and examines the query data from another anycast system to test those inferences.