Data-driven Approaches to Edge Caching

Guangyu Li,Qiang Shen,Yong Liu,Houwei Cao,Zifa Han,Feng Li,Jin Li

DOI: https://doi.org/10.1145/3229574.3229582

2018-01-01

Abstract:Content caching at network edge is a promising solution for serving emerging high-throughput low-delay applications, such as virtual reality, augmented reality and Internet-of-Things. The traditional caching algorithms need to adapt to the edge networking environment since old traffic assumptions may no longer hold. Meanwhile, user/group content interest as a new important element should be considered to improve the caching performance. In this work, we propose two novel caching strategies that mine user/group interests to improve caching performance at network edge. The static user-group interest patterns are handled by the Matrix Factorization method and the temporal content request patterns are handled by the Least-Recently-Used or Nearest-Neighbor algorithms. Through empirical experiments with a large-scale real IPTV user traces, we demonstrate that the proposed caching algorithms outperform the existing caching algorithms and approach the caching performance upper bound in the large cache size regime. Leveraging on offline computation, we can limit the online computation cost and achieve good caching performance in realtime.