Probabilistic Caching Based on Maximum Distance Separable Code in a User-Centric Clustered Cache-Aided Wireless Network

Abstract

We investigate how the maximum-distance separable (MDS) coding can be incorporated into probabilistic caching to utilize the limited storage space efficiently. In a user-centric clustered wireless network, each caching helper node probabilistically caches a segment of MDS coded sequence of each file. The segment size is optimized to maximize the cache hit probability or successful file retrieval probability. We reveal that the best way of storing files is determined by the condition whether the average amount of MDS coded information stored for the requested file within a user's cluster exceeds the amount required for file retrieval or not. In terms of the cache hit probability maximization, if the condition is not fulfilled, it is proved that storing the complete file with a low probability is optimal. Otherwise, storing either a segment as small as possible with a high probability or a complete file with a low probability, according to a given environment, is shown to be desirable. We also analyze the successful retrieval probability, which accounts for both a cache hit event and successful transmissions from multiple caching helper nodes. Since the successful retrieval probability is in an intractable form, to find the desirable segment size, the theoretically driven algorithms with low search complexity are developed.