Coded Caching with Heterogeneous File Demand Sets — The Insufficiency of Selfish Coded Caching

Abstract

This work falls under the broad setting of coded caching with user-dependent file popularity and average-rate capacity analysis. In general, the exact capacity characterization with user-dependent file popularity remains an open problem. For example, user 1 may be interested in files 1 and 2 with probabilities 0.6 and 0.4, respectively, while user 2 may be interested in only files 2, and 3 with probabilities 1/3 and 2/3, respectively, but not interested in file 1 at all. An optimal scheme needs to carefully balance the conflicting interests under the given probabilistic weights. Motivated by this fundamental but intrinsically difficult problem, this work studies the following simplified setting: Each user k is associated with a file demand set (FDS) Θk; each file in Θk is equally desired by user k with probability $\frac{1}{{\left| {{\Theta _k}} \right|}}$; and files outside Θk is not desired at all. Different users may have different Θk1 ≠ Θk2, which reflects the user-dependent file popularity. Various capacity results have been derived (mostly for the cases of K = 2 users). One surprising byproduct is a proof showing that selfish coded caching is insufficient to achieve the capacity. That is, in an optimal coded caching scheme, a user sometimes has to cache the files of which he/she has zero interests.