Deep Learning-Enabled File Popularity-Aware Caching Replacement for Satellite-Integrated Content-Centric Networks

Abstract

In recent years, satellite-integrated content centric networking (SCCN) has become an important solution for the future network with excellent bandwidth savings and file distribution capability in a wide range by the intrinsic caching function. At present, however, the high-speed changes of satellite network topology and coverage make it difficult to predict the file popularity of SCCN network, resulting in a low timeliness. This will lead to low node cache efficiency and bad data distribution performance. To address this issue, in this article, we have proposed a deep learning-enabled file popularity-aware caching replacement mechanism to achieve efficient file distribution in SCCN. In the proposed mechanism, we have developed a virtual location division scheme to keep the return path of content data invariable by remapping the time-varying topology of network into a static topology with virtual nodes. Furthermore, we have put forward a minimum delay file-caching set algorithm to predict the popularity of files in the proposed SCCN via a well-designed deep learning framework, which can find those high-popularity files most worthy of caching. The simulation results verified the proposed method can obviously degrade the access delay of all users and the cache hit ratio of satellite nodes, compared with current strategies, i.e., cache everything everywhere with least recently used, probCache, content-aware placement and discovery, and replacement (APDR), respectively.