Dimensioning backbone networks for multi-site data centers: Exploiting anycast routing for resilience

In the current era of big data, applications increasingly rely on powerful computing infrastructure residing in large data centers (DCs), often adopting cloud computing technology. Clearly, this necessitates efficient and resilient networking infrastructure to connect the users of these applications with the data centers hosting them. In this paper, we focus on backbone network infrastructure on large geographical scales (i.e., the so-called wide area networks), which typically adopts optical network technology. In particular, we study the problem of dimensioning such backbone networks: what bandwidth should each of the links provide for the traffic, originating at known sources, to reach the data centers? And possibly even: how many such DCs should we deploy, and at what locations? More concretely, we summarize our recent work that essentially addresses the following fundamental research questions: (1) Does the anycast routing strategy influence the amount of required network resources? (2) Can we exploit anycast routing for resilience purposes, i.e., relocate to a different DC under failure conditions, to reduce resource capacity requirements? (3) Is it advantageous to change anycast request destinations from one DC location to the other, from one time period to the next, if service requests vary over time?