Resource Allocation And Decision Mechanism In Stochastic And Dynamic Wireless Network

In order to integrate Internet and mobile networks,network,computing,storage functions are placed at the edge of wireless network.Therefore,enhancing the capability of wireless network edge will be an important research field to maximize the potential of 5G.However,due to the random movement of users,how to consider the impact of these stochastic and uncertainties on resource allocation and decision-making is still a challenge.In infrastructure level,due to the random movement,the connection between mobile user and edge server is uncertain.Therefore,a Multi-Stage Stochastic Programming(MSSP)methodology-based decision-making is proposed for offloading,computing resource allocation and migration decisions sequentially.The objective of MSSP is to minimize the expected cost function while considering the all possible uncertain scenarios.Moreover,due to the high computational complexity of MSSP,a Sample Average Approximation(SAA)algorithm is used to obtain a near-optimal decision for the proposed MSSP problem(SAA-MSSP),which samples scenarios independently from all scenarios to obtain a sample average function quickly and effectively.Compared with the conventional offloading policy in static or quasi-static environment and mean-based offloading policy,simulation results reveal that the proposed MSSP and SAA-MSSP decisions can achieve the better offloading performance in term of the total cost and energy consumption,the gap between approximation decision and optimal decision is reasonable.In mobile edge infrastructure level,MEC orchestrator should determine multidimensional resource allocation for Virtual Network Request(VNR),and map VNR on the substrate infrastructure.However,due to random arrival of requests and dynamic time-varying resource,greedy mapping cannot always be optimal,especially in the overloaded infrastructure.Considering the impact of dynamic environment on mapping,two opportunistic virtualization strategies are proposed,which allows VNR to decide whether to map or to wait on each stage.In order to cope with the dynamic environment,the multi-dimensional resource allocation problem is formulated as a Markov Decision Process(MDP)problem to obtain the optimal virtualization strategy maximizing the overall reward,and MDP-based opportunistic virtualization strategy(M-OVS)is designed in a centralized way.Furthermore,considering the exponential complexity to solve MDP problem,the buyer-seller game theoretic approach is adopted to motivate VNR to maximize its own reward in a distributed manner(G-OVS).Numerical results show that the proposed strategies could make the reasonable virtualization decision effectively and steadily,and they outperform the baseline scheme that maps greedily in terms of system reward,overall cost,resource utilization ratio and virtualization failure probability.