Research On Workload Scheduling And Service Caching Strategy In Ultra-Dense Networks

With the rapid development of emerging mobile applications such as augmented/virtual reality,Internet of Vehicles and autonomous driving,mobile terminals cannot meet the requirements of mobile applications for computing and storage resources due to their own capabilities.Mobile Edge Computing(MEC)can effectively process mobile data generated at the edge of the network by deploying computing,storage and network services at the edge of the network,meeting the requirements of mobile applications for Quality of Service(Qo S).However,the explosive growth of mobile data and massive device connections have brought huge challenges to 5G networks.Ultra-Dense Network(UDN),as a key technology in 5G,increases the number of connections of mobile devices in the network through dense deployment of low-power small base stations and hotspots,provides good access services for mobile devices.The explosive growth of mobile data traffic and the demand for massive device connections.Therefore,by deploying mobile edge computing servers(MEC-Enabled Small Cell Base Station,MEC-SBS)on micro base stations in ultra-dense networks,it is possible to effectively process edge data,reduce backhaul network data transmission and improve end-user Qo S.However,in the ultra-dense network,with the dense deployment of MEC-SBS,the network scale gradually becomes larger,coupled with the small coverage of MEC-SBS and limited resources,the computing workload and application service requirements on MEC-SBS are easily affected by space and time.Due to the influence of factors such as user movement,the computing workload and application service demand on it change dynamically and distribute unevenly.How to effectively schedule the computing workload on MEC-SBS and application service caching to improve user service quality and computing workload processing efficiency is a challenging problem.The main research contents of this paper are as follows:(1)The MEC-SBS division of cooperative clusters is used to solve the large-scale workload scheduling problem in ultra-dense networks.By using the clustering algorithm,the MEC-SBS in the system is divided into multiple non-overlapping computing cooperative clusters,thereby realizing transforming the large-scale MEC-SBS computing workload scheduling problem into the small-scale MEC-SBS computing workload scheduling problem in the computing cooperative cluster.Each computing cooperative cluster realizes the computing workload scheduling in the cluster through distributed parallel execution.Considering that the computing workload arrival information on MEC-SBS is not known,the DDPG algorithm based on deep reinforcement learning is adopted,and the optimal cluster computing workload scheduling strategy is made according to the MEC-SBS computing workload information in the cluster,which guarantees the energy of MEC-SBS under the condition of consumption,the average service delay of computing tasks in the cluster is minimized.In order to deal with the problem of unbalanced computing workload among cooperative clusters,a semi-dynamic K-Means cooperative cluster clustering algorithm based on load balancing is proposed,which re-clusters the computing task overload cluster and its neighbor clusters into cooperative clusters,and realizes workload balancing among cooperative clusters.(2)Aiming at the problem that the application service requirements on the MEC-SBS in the system change due to factors such as time,space,and user service requests,this paper models the application service request information based on the n-order Markov chain and designs application service caching update algorithm based on service popularity.The algorithm based on the application service request information in the first n time slots calculates the popularity of each application service in the current time slot,and calculates the application service cache in the current time slot,which improves the application service in the system caching update accuracy and reduces update complexity.Considering that application service requests from the same area in the network usually have a certain similarity,this paper proposes a collaborative cluster division algorithm based on the similarity of application services,which divides the MEC-SBS with high similarity of application services in the system into the same collaborative cluster,using the abovementioned workload scheduling algorithm based on DDPG,the optimal load scheduling strategy is made according to the application services cached by MEC-SBS and the request information of various types of computing tasks,so as to maximize the utility of the service caching.