Caching over-the-top services, the Netflix case

Abstract

Over-the-top (OTT) traffic represents around 67% of the downstream Internet data volume during peak periods in the US. OTT content providers have been following the strategy of deploying caching servers within the ISPs' own networks. This work presents a method for deciding where these caches for OTT services can be conveniently deployed to provide significant traffic offload to ISPs' back-haul links. The procedure followed is to find suitable caching locations that minimize the maximum link traffic load. The problem is defined and solved as a Link Load Balanced Capacitated Facility Location Problem (LLB-CFL). The solution search processes are implemented based on Genetic Algorithms (GA), designing genetic operators highly targeted towards this specific problem. The proposed methods are applied to a case study focusing on the demand and cache specifications of Netflix, and framed into a real geographical area, the island of Bornholm, Denmark. The results cover a thorough analysis of the cost-fitness trade-off of the search method and an evaluation of both cache and gateway traffic flowing in the network in relation to the solutions found using different network topologies.