Decentralized coded caching for shared caches using erasure coding

Abstract

In a client–server architecture data traffic varies through time. It exhibits congestion of network resources at peak hours while the network remains rather under-utilized during the off-peak hours. One of the potential techniques to smooth out this data traffic over time is to load part of the data into cache memories distributed across the network. In the classical cache framework, each user is associated to a cache and the caching process is centrally coordinated. However, for large networks, decentralized caching is more practical since decentralized caching does not require the knowledge of all the active users in the network during the prefetching phase which is a necessity for centralized caching. Furthermore, with an aim to reduce cache under-utilization in a network, caches are allowed to serve multiple clients. Such a scheme in which multiple users are allowed to share a cache is referred to as the shared caching scheme. Shared caching scheme has been explored in both centralized and decentralized setting with centralized scheme giving a superior performance. To improve the performance of the decentralized caching scheme we implement precoding at the server’s end. Precoding is done by encoding the data stored in the server using a Maximum Distance Separable (MDS) erasure code. The coded caching framework in this paper is denoted as MDS precoded decentralized shared caching. We propose a delivery scheme and derive a delivery rate for the proposed scheme. Using techniques from index coding, a lower bound is obtained for all MDS precoded decentralized shared caching linear delivery schemes delivering certain contents. The proposed lower bound is shown to meet the proposed delivery rate. The achievable delivery rate proposed in this paper is also shown to match with the rate obtained for the optimal centralized shared caching scheme for small and large cache memory sizes.