Codes for a distributed caching based Video-on-Demand system

We study the role of codes in the optimization and design of a large-scale Video-on-Demand (VoD) system based on distributed caching, that we have architected and built at Berkeley. We show how network codes can convert a combinatorial problem into a tractable one, and enable a fully distributed algorithm that jointly optimizes the three-fold problem of cache content placement, cache-to-users topology selection, and cache-to-users rate-allocation. While a description of the general VoD system optimization and design can be found in [1], this paper focuses on the critical role of codes in enabling our VoD system. Specifically, we motivate and describe a specific class of network codes, called DRESS codes, that offer desirable tradeoffs between cache-to-user and cache-to-cache communication aspects of the problem needed to sustain a scalable VoD system.