Caching in Wireless Networks

TL;DR

This paper characterizes the fundamental limits of content caching in large wireless networks, providing bounds on capacity regions and revealing the importance of joint cache selection and load balancing.

Contribution

It offers the first information-theoretic scaling characterization of caching capacity in wireless networks, with a novel scheme separating cache selection and coding.

Findings

Inner and outer bounds on caching capacity region in high path-loss regime

Separation of cache selection and channel coding is order-optimal

Nearest-neighbor cache selection can be arbitrarily inefficient

Abstract

We consider the problem of delivering content cached in a wireless network of n nodes randomly located on a square of area n. The network performance is described by the n2^n-dimensional caching capacity region of the wireless network. We provide an inner bound on this caching capacity region, and, in the high path-loss regime, a matching (in the scaling sense) outer bound. For large path-loss exponent, this provides an information-theoretic scaling characterization of the entire caching capacity region. The proposed communication scheme achieving the inner bound shows that the problems of cache selection and channel coding can be solved separately without loss of order-optimality. On the other hand, our results show that the common architecture of nearest-neighbor cache selection can be arbitrarily bad, implying that cache selection and load balancing need to be performed jointly.