Content delivery via multi-server coded caching in linear networks

Abstract

We consider a content delivery network where multiple servers are connected to multiple cache-enabled clients. Clients request their corresponding contents from the servers and servers collaboratively transmit packets to fulfill all the requests. It is assumed that some contents are stored in the caches in off-peak time of the network without knowing the actual requests, the so called cache content placement phase. The goal is to minimize the worst case delay in the content delivery phase. Considering a random linear network, we propose a coding strategy which exploits servers' multiplexing gains as well as caches' global and local coding gains. The main idea in our coding scheme is to expand the number of users benefiting from a single packet by using zero-forcing techniques. This results in an increase in multi-casting gain which in turn provides faster delivery of the contents to the users. In addition, we show that our scheme is optimal for a certain regime of parameters.