Multi-dimensional Resource Management In Mobile Edge Computing Networks

With the rapid popularity of User Equipments(UEs)and the fast development of the Internet of Things(IoT),it is expected that more and more computational intensive mobile applications,eg.,face recognition and virtual realization,etc.,will be running on the UEs.These applications are typically characterized by huge energy consumption,requiring powerful processing capabilities and extremely short latencies.However,UEs are usually resourceconstrained,possessing limited processing capabilities and inadequate battery lives,making them unable to cope with the attendant challenges in task processing or energy supply.At the same time,many new bandwidth-demanding services such as mobile video and online social media,etc.,have brought exponentially growth in mobile data traffic,making the backhaul link become a new bottleneck in affording the huge traffic load.Mobile Cloud Computing(MCC)has been considered as a promising solution to address UEs' restrictions by offloading the applications to powerful cloud center.However,the latency caused by transferring data to a remote cloud server is often unacceptable for some latency sensitive applications.Mobile Edge Computing(MEC)or fog computing has been proposed as an effective supplement to MCC by providing IT and cloud computing capabilities within the Radio Access Networks(RANs),which can enable processing,caching and networking services to the nearby UEs.Consequently,MEC can not only reduce the energy consumption for UEs and decrease the processing latency for applications,but also can relieve the traffic burden for backhaul links.However,most of the current literatures about mixed cloud and fog computing just stayed at proposing a system architecture,without modeling and further problem solving.In addition,most works did not consider to combine computation offloading and Internet content caching together in MEC paradigm,so as to provide more satisfactory services.In this paper,we devote to studying the key problems involved in MEC systems,including computation offloading,resource allocation,and content caching.Specifically,the main contents of this paper include:(i)We perform joint optimization of computation offloading,computational resource allocation,wireless bandwidth allocation and transmit power control in a heterogeneous MEC system,which comprise a cloud server,a WiFi Access Point(AP)based MEC server,and multiple UEs.We intend to minimize the cost of weighted energy consumption and processing delay,while guaranteeing the maximum delay tolerance of tasks and ensuring the min-max fairness among UEs.Through Semi-Definite Relaxation(SDR),we first obtain the binary offloading decision for each UE,and then we get the MEC server's computational resource allocation based on bisection method.Since joint transmit power and wireless bandwidth allocation is a non-smooth non-convex optimization problem,we transform it into a convex optimization problem by employing fractional programming and slacking.Thus the transmit power control and wireless bandwidth allocation can be obtained by Lagrangian dual decomposition,sub-gradient projection and one-dimensional search.Simulation results verify the convergence of our iterative algorithms,and exhibit the proposed algorithms perform well in terms of fairness,delay,energy consumption and the number of benefited UEs(a UE is called a benefited UE,which means its cost can be reduced compared with local execution).(ii)We perform joint optimization of computation offloading,computational resource allocation,RB allocation,and transmit power control in a heterogenous MEC system,which is composed by a remote cloud server,an LTE-A small cell Base Station(BS)based MEC server,and multiple UEs.We perform optimization so as to enable computational-intensive and low-latency-demanded applications in UEs.We propose a Binary tailored Fireworks Algorithm(BFA),based on which we propose an iterative algorithm framework to solve the constructed problem.In this framework,we first obtain the UEs' task offloading decision using BFA,and then get the computational resource allocation based on bisection method.The difficulty of the problem lies in the limitation of radio resource allocation in the LTE-A uplink,i.e.,one RB can be assigned to only one UE,and multiple RBs allocated to one UE must be adjacent.To solve this problem,we call several adjacent RBs as an RB pattern(which is shorted as ”pattern” for notation simplicity),and then we allocate feasible patterns instead of RB allocation.Applying continuous variable relaxation and Lagrangian dual decomposition,each pattern is assigned to an optimal UE,and the transmit power control is also performed on each RB within an pattern,however,one UE may be assigned with multiple feasible patterns and may violate the the requirement that one UE can only be assigned with multiple adjacent RBs.For this purpose,we propose a heuristic algorithm for each UE to pick out the optimal pattern from its available patterns,while guaranteeing the exclusiveness of RB allocation and high RB utilization.Simulation results verify the convergence of our iterative algorithms,and demonstrate that the proposed joint optimization algorithm can obtain better delay performance under different parameter settings.(iii)We study the problem of joint computation offloading and content caching in an MEC network,and construct the problem as a stochastic optimization problem.Under the premise of guaranteeing the stability UEs' data queues and caching queues,we perform the joint optimization of the offloading decisions and content caching decisions,wireless rate allocation,computation resource allocation,and caching resource allocation,in order to maximize the operator's long-term averaged economic benefits.Adopting Lyapunov optimization theory,we develop an online algorithm for dynamic joint computation offloading,resource allocation and content caching,where the original formulated problem is first reconstructed into the each-time-slot optimization,and then decoupled into two independent subproblems and are finally solved by 0-1 integer programming and linear programming.Simulation results show that the proposed algorithms can obtain more economic profit with the change of system parameters and operator's pricing strategies,and also demonstrate that the control parameter V can make a good balance between the operator's economic profit and the processing delay of tasks.