Abstract

Mobile network performance and user Quality of Experience have been negatively affected by the explosion of mobile data traffic. This paper proposes mobile edge caching to alleviate the problem. Recent research has focused on local caching at the wireless edge, as motivated by the 80/20 rule regarding content popularity. By caching popular contents at base stations (BSs) closer to users, backhaul congestion and content access latency can be dramatically reduced. To address the limited storage size of BSs in the context of the massive amount of available content, an algorithm optimizing cooperative caching has been highlighted. Contents requested by mobile users that cannot be obtained locally could be transferred by cooperative BSs. In this paper, we propose a cooperative caching algorithm based on BS content access patterns. We use tensor decompositions with distance constraint to analyze interaction between users, contents and base stations. Thus, BSs with small geographical distances and similar content access patterns constitute a cooperative caching domain. The distributed content placement and delivery algorithm is optimized based on simultaneous consideration of the caching hit ratio and cooperative cost. Simulation results based on a real dataset of usage detail records demonstrate the superior performance and promising practical gains in caching of the proposed caching method compared to user clustering and BS clustering.

中文翻译：

---

基于网络边缘内容访问模式的协作缓存和交付算法

摘要

移动网络性能和用户体验质量已受到移动数据流量激增的负面影响。本文提出了移动边缘缓存来缓解该问题。受有关内容受欢迎程度的80/20规则的启发，最近的研究集中在无线边缘的本地缓存。通过将流行内容缓存在离用户更近的基站（BS）上，可以大大减少回程拥塞和内容访问延迟。为了解决在大量可用内容的情况下BS有限的存储大小，已经提出了一种优化协作缓存的算法。移动用户请求的本地无法获得的内容可以由协作BS传输。在本文中，我们提出了一种基于BS内容访问模式的协作缓存算法。我们使用具有距离约束的张量分解来分析用户，内容和基站之间的交互。因此，具有小的地理距离和相似的内容访问模式的BS构成了协作缓存域。基于同时考虑缓存命中率和协作成本来优化分布式内容放置和交付算法。基于使用情况详细记录的真实数据集的仿真结果表明，与用户集群和BS集群相比，所提出的缓存方法在缓存方面具有出色的性能和可观的实际收益。基于同时考虑缓存命中率和协作成本来优化分布式内容放置和交付算法。基于使用情况详细记录的真实数据集的仿真结果表明，与用户集群和BS集群相比，所提出的缓存方法在缓存方面具有出色的性能和可观的实际收益。基于同时考虑缓存命中率和协作成本来优化分布式内容放置和交付算法。基于使用情况详细记录的真实数据集的仿真结果表明，与用户集群和BS集群相比，所提出的缓存方法在缓存方面具有出色的性能和可观的实际收益。