Research On Task Caching In Edge Cloud With Multi-Armed Bandit

In recent years,with the rapid growth of the number of mobile terminal devices,the traffic load of cloud computing network has increased significantly,resulting in higher task execution time.Meanwhile,with the increasing intelligence of mobile applications,intelligent mobile devices cannot meet the computing and latency requirements of emerging mobile applications such as virtual reality and augmented reality,so these mobile applications can only hope for new network architectures.Many researchers have proposed the edge cloud computing architecture.The edge cloud server can cache some tasks that were originally in the remote cloud.The edge cloud caching architecture is able to slow down network jamming and reduce task execution time because of task requests of the smart mobile device can be executed in the near place.The task caching solution can meet the low latency requirements of emerging mobile applications.Considering the limited cache capacity of the edge cloud server,in order to select some tasks to be cached into the edge cloud server.We propose an adaptive task caching strategy called UCB-AC(Upper Confidence Bound-Adaptive Caching)based on Multi-Armed Bandit model.The UCB-AC task caching algorithm not only can learn the mobile users' request pattern online,but also dynamically adjust the caching policy according to the amount of users' task.We use mathematical methods to prove the upper bound of cumulative expected learning regret of the algorithm,which indicates that the UCB-AC algorithm can reduce the execution time of the task with a limited learning cost.In order to evaluate the effectiveness of the UCB-AC algorithm,we use average task duration,expected learning regret,and cumulative expected learning regret as the evaluation indicators.The contrast experiments are carried out in the edge cloud caching architecture simulation model.The experimental results show that compared with the traditional Multi-Armed Bandit task caching strategies,UCB-AC algorithm can minimize the execution time and learning regret of the task.