Abstract

In a first part of the talk, we consider the so-called live streaming problem, which consists in broadcasting a stream of information to a collection of network nodes, and we attempt to achieve this objective while minimizing network costs. To this end, we propose the so-called “Implicit primal-dual” scheme, whose simplicity makes it an interesting practical candidate. We characterize its behaviour at a fluid time scale, and obtain desirable properties when the scheme is combined with random linear coding. In a second part of the talk, we consider the so-called Video-on-demand problem, in which users request access to content items at arbitrary random times. In this context, we consider a simple user policy for managing which content to keep. We use the general framework of loss networks to model the resulting performance, and obtain optimality results on the ability of our scheme to enable users to serve each others’ requests.