Abstract: In Augmented Reality(AR) applications,user requests that are geographically close to each other might be similar or identical,leading to repeated task execution.To address the problem,this paper designs a cache system for computation tasks based on redundant task reduction.Task cache is designed for the edge nodes,making the edge servers maintain the global cache in a self-organized way,and the client request latency,user trajectory,node deployment and total latency are mathematically modeled.On this basis,this paper studies the computation resource deployment of edge servers in base stations,optimizing the average request delay under a given deployment cost.The deployment problem is simplified to an integer nonlinear programming problem,and then two algorithms are presented:Integral Delay Minimization(IDM) for medium and small scale scenarios,and Large-scale Delay Minimization(LDM) for large scale scenarios.Experimental results show that the difference between the average delay of the IDM algorithm and that of the optimal solution is only 5.85%,which means the proposed algorithm has a very good approximation effect on the optimal solution. Compared with the IDM algorithm,the LDM algorithm reduces running time by 98.15% at the expense of 9.20% longer average delay,greatly reducing the running cost.

Key words: Augmented Reality(AR), edge computing, redundant task, dynamic programming, clustering

摘要： 在增强现实应用中，距离较近的多个用户请求很可能是相似或者相同的，从而导致同样的计算任务被重复执行。针对该问题，设计基于冗余任务消减的计算任务缓存系统。通过在边缘节点设计任务缓存，使边缘服务器以自组织方式维护全局缓存。对客户端请求时延、用户轨迹、节点部署和总时延进行建模，基于此研究基站上边缘服务器的计算资源部署问题，在给定总的部署代价下优化平均请求时延，并将该问题转化为整数非线性规划问题，设计针对中小规模场景的IDM算法和针对大规模场景的LDM算法。实验结果表明：IDM算法的平均时延与参考最优解仅相差5.85%，对最优解具有较好的逼近效果；LDM算法在牺牲9.20%平均时延的情况下，相比于IDM算法运行时间缩短98.15%，大幅减少了运行开销。

关键词: 增强现实, 边缘计算, 冗余任务, 动态规划, 聚类

CLC Number: 

• TP393