12 is enough
The paper authored by Ricardo de O. Schmidt, John Heidemann and Jan Harm Kuipers addresses the impact of the size of an anycast deployment, and how the anycast sites are distributed and connected to the Internet, on the performance of the service. The paper shows that around 12 anycast sites are enough to provide good connectivity and low latency to the majority of the Internet (having the RIPE Atlas as observation points). It also shows that much larger deployments can result in services with very low latency, provided that the anycast sites are well distributed geographically and, most importantly, well connected to the neighboring Autonomous Systems in their respective regions.
Some DNS and CDN operators are using the results of this work to help them understanding the behavior of their anycast deployments, and to define strategies for service expansion. SIDN will be using the results to further optimize their DNS anycast service for .nl.
Also, the findings are a starting point of the second phase of the SAND project (SAND-2), which will again be a collaboration of the University of Twente, SIDN Labs and NLnet Labs. The work presented in this paper was the first step towards understanding the relationship between anycast and Internet routing (as defined by the BGP protocol), which is one of SAND-2’s deliverables.
Jan Harm Kuipers, currently doing his master’s thesis at SIDN Labs, presented the paper on 31st March 2017, among many other papers from top institutions around the globe that addressed a wide range of networking problems using measurements as main approach.
Oh, and I will be joining the SIDN Labs team as a Research Engineer on 1st May 2017 :-)