An Online Coded Caching Scheme Without File Splitting

Coded caching has proven to be an effective technique to alleviate bottlenecks in wireless networks. By preemptively storing some content at the end users during off-peak hours, it creates multicasting opportunities that can be leveraged to reduce the peak rate when the network is congested. However, in a practical setting it is often necessary to update the cache contents dynamically based on the current request distributions and network configuration. This paper proposes a decentralized algorithm for updating the cache contents in coded caching systems. The proposed algorithm stores and transmits whole files, without splitting them, and users only rely on overheard messages to update their caches. The asymptotic performance of the proposed algorithm is derived analytically and simulations show that it outperforms the Least Frequently Used (LFU) eviction policy and the uniform memory allocation policy for a wide range of file popularity distributions. Additionally, they prove that the advantages of the proposed algorithm are not limited to a single cherry-picked scenario, but rather hold for various number of users, files, and cache sizes.