Collaborative Mechanism And Optimization Of Communication,caching And Computing In Mobile Edge Computing Networks

With the rapid development of mobile Internet and the continuous intelligence of mobile terminals,data-computing-intensive and ultra-low latency demands for mobile services are exploding,which leads to the increasing shortage of wireless communication bandwidth.The traditional connection-oriented mobile network architecture cannot effectively meet the demand for diversified mobile services due to the lack of content awareness.By fully mining the rich computing and caching resources at the network edge nodes,mobile edge computing(MEC)network has developed the ability of content perception,and thus is regarded as one of the key technologies that can effectively solve the bottleneck of massive information transmission.In this thesis,according to the mobile service type from simple to complex,the collaborative mechanism of communication,caching and computing in MEC network is built,and the joint optimal caching and computing policy at the mobile devices is designed to minimize the bandwidth cost for the data-intensive content distribution services,cloud-input and bidirectional input computation tasks,respectively.The specific research contents and contributions of this thesis are summarized as below.· Firstly,for the delivery of data-intensive content distribution services,the pushing policy at the base station(BS)and caching policy at the mobile devices are jointly designed to maximize the wireless bandwidth utilization.In particular,a multi-user MEC system with caching and multicast capabilities is considered.Aimed at maximizing the bandwidth utilization,the joint pushing and caching optimization problem is formulated as a Markov decision process.Based on the Bellman equation,it is illustrated how to get the best balance between the current transmission cost and the future average transmission cost.Based on the coupling and interchange arguments,it is also proved that the optimal average transmission cost decreases with the cache size.When the file request probability model can be predicted in advance,a low complexity decentralized policy(LDP)is designed via linear approximation of the value function and transforming the challenging dynamic discrete optimization problem into a difference of convex problem.When the file request probability model is unavailable in advance,an online distributed algorithm based on Q-learning is proposed to implement LDP.The simulation results show that the proposed algorithm approximates the optimal performance under small system parameters,and exhibits better performance in terms of bandwidth gain under general system parameters.· Secondly,for the delivery of cloud-input computation task,the caching and computing policy at the mobile device is jointly designed to minimize the wireless bandwidth cost.In particular,a single-device MEC network with caching and computing capabilities is considered.The mobile device can not only proactively cache the input and output data of some computation tasks,but also compute some tasks locally.Under the constraints of service latency,local cache and computation resources,the caching and computing policy at the mobile device is jointly optimized.When the system parameters are homogeneous,the closed-form expressions for the optimal joint policy and the corresponding minimum wireless bandwidth requirement are derived,thereby theoretically revealing the tradeoff among communication,computing and caching.According to the relationship between the transmission rate under the MEC downloading policy and that under the local computing policy without local caching,the system feasible region is divided into the local computing limited region and MEC downloading limited region.In each region,the research results show that the joint utilization of the caching and computing resources at the mobile devices facilitates the bandwidth gain achievement.When the system parameters are heterogeneous,the original optimization problem is transformed into a linearly constrained indefinite quadratic problem,and then the local optimum is obtained based on the concave convex method.In addition,take the mobile virtual reality(VR)delivery as a use case and a novel MEC-based mobile VR delivery framework is designed.The numerical results show that the proposed mobile VR delivery framework can significantly reduce the wireless communication bandwidth while meeting the low latency requirement.· Then,for the delivery of cloud-input computation task,the caching and computing policy at multiple mobile devices is jointly designed under multicast opportunity to minimize the wireless bandwidth cost.In particular,a novel MEC system is proposed,in which the MEC server caches the input and output data of all computation tasks,and communicates with multiple mobile devices via a shared wireless link.Each task request from each mobile device can be served via local output caching,local computing with input caching,local computing without local caching or MEC downloading,and each service method incurs a certain bandwidth demand for multicast link.Under the constraints of local latency,cache and computation resources as well as multicast transmission,the joint caching and computing policy is established to minimize the average transmission bandwidth on the multicast link.In order to solve the problem that the priori knowledge of user request probability is difficult to predict,the expectation of bandwidth cost is approximated by sampling method.When the output data size is smaller than the input data size,it is proved that the wireless bandwidth gain only comes from the local cache and has nothing to do with the local computation.Based on this property,the original optimization problem is transformed into the monotonically nonincreasing submodular minimization problem under matroid constraints,and the optimal joint policy can be obtained by strongly polynomial algorithm of Schrijver.Otherwise,in order to reduce the computation complexity,a low-complexity high-performance algorithm is proposed by combining the convex concave procedure algorithm and alternative direction method of multipliers,and it is proved that the algorithm can converge to a local minimum.In addition,when the system parameters are homogeneous,the wireless bandwidth gains from the computing and caching resources at the mobile devices as well as the multicast are theoretically revealed.The numerical results show that the joint utilization of computing and caching resources on mobile devices and multicast can greatly save wireless communication bandwidth.· In the end,for the delivery of bidirectional-input computation task,the caching and computing policy at the mobile device is jointly designed to minimize the wireless bandwidth cost.In particular,a single-user MEC system with caching and computing capabilities is considered and a novel bidirectional-input computation task model is proposed,in which the input data of each computation task consists of the locally generated data in real time and the cacheable data proactively generated in the cloud.Each computation task can be served via local computing with local caching,local computing without local caching or MEC computing,each of which incurs a certain wireless bandwidth cost.Under the constraints of service latency,local storage and average computation power,the caching and computing policy at the mobile device is jointly optimized to minimize the average bandwidth cost.The closed-form expressions for the optimal strategy and the minimum bandwidth are derived,and the interaction among communication,caching and computing for bidirectional input computation task is discussed from both theoretical and numerical results.