A scalable model for multi-period virtual network mapping for resilient multi-site data centers

摘要: In the currently dominant cloud computing paradigm, applications are being served in data centers (DCs) which connected to high capacity optical networks. For cost efficiency reasons, both DC and network domains, virtualization of physical hardware is exploited. a DC, it means that multiple so-called virtual machines (VMs) hosted on same server. Similarly, partitioned into separate networks, thus providing isolation between distinct operators (VNOs). Thus, problem mapping arises: how decide resources allocate for particular network? this paper, we study context with sites. This introduces additional flexibility, due anycast routing principle: have freedom at what location serve application. We can exploit choice minimize required when solving problem. paper builds our earlier work solves resilient optimally decides center resources, considering time-varying traffic conditions protecting against possible failures resources. consider VNO resilience scheme: rerouting under failure provided layer. To resource requirements, allow reuse resources: scenarios assumed not occur simultaneously. Since also protect failures, backup account synchronization primary DCs. deal time variations volume geographical pattern application traffic, investigate potential benefits (in terms overall requirements) reconfiguring from one period next. Previously, developed model solve multi-period case step time: given fore t, determine (possibly changed) t + 1. Compared previous work, now (i) define truly path formulation exploiting column generation, (ii) demonstrate its scalability nation-wide hourly varying traffic.