移动边缘计算中基于内容流行度的深度强化学习缓存机制

随着5G商用的推进,涌现出大量依赖高速率、低时延的新应用,混合现实(Mixed Reality,MR)就是其中之一。考虑到从中心云传输服务内容到MR设备会带来很大时延和能耗问题,引入移动边缘计算(Mobile Edge Computing,MEC)技术,通过在MEC服务器上缓存用户的预渲染环境帧,以减少延迟和能耗。针对MEC服务器上有限的缓存资源,提出了一种基于内容流行度的深度强化学习(Deep Reinforcement Learning,DRL)方法来做缓存决策,并构造一个新的效用函数来衡量缓存方案的性能;仿真结果表明基于所提算法得到的缓存决策能使目标效用函数达到最大值。

A Content Popularity Based Caching Scheme with Deep Reinforcement Learning for Mobile Edge Computing

With the advancement of 5 G commercial use, a large number of new applications that rely on high speed and low latency have emerged, e.g.,Mixed Reality(MR).Considering the transmission of service content from the central cloud to the MR device will bring great delay and energy consumption, the Mobile Edge Computing(MEC) technology has been introduced.It can reduce latency and energy consumption by caching the users’ pre-rendered environment frames on the MEC server.With the limited cache resources on the MEC server, a content caching scheme based Deep Reinforcement Learning(DRL) method was proposed to make caching decisions.Then, a new utility function was proposed to measure the performance of the caching scheme, and the proposed scheme was simulated and verified.