Centralized caching in two-layer networks: Algorithms and limits
Lin Zhang, Zhao Wang, Ming Bo Xiao, Gang Wu, Shaoqian Li · 2016
The problem of the centralized caching is studied in a two-layer network. The first layer of the network is constructed by a server and K1helpers, and the second layer consists of K1orthogonal sub-networks, each of which contains a helper and K2users. The pioneer caching design in the two-layer network is to directly apply the Maddah-Ali & Niesen (MAU) centralized caching [1] into individual layers, such that single-layer multicast opportunities (SMO) are deployed. In this paper, a joint caching (JC) algorithm is developed by exploiting both the SMO and the correlations of caching contents across two layers, namely, cross-layer storage correlations (CSC). Furthermore, cross-layer multicast opportunities (CMO) can also be created by applying the MAU scheme between the server and users. In order to simultaneously obtain the caching gains from SMO, CSC, and CMO, a hybrid caching scheme is proposed and demonstrated to be order-optimal when the storage sizes at both helpers and users are limited. In other words, the achievable rate region lies within a constant multiplicative and additive gap of the information-theoretic bounds. In particular, the multiplicative and additive factors can be carefully characterized to be 1/48 and 4, respectively.