Research On Computation Offloading Strategies In Mobile Edge Computing

The emergence of resource-intensive and latency-sensitive applications such as smart factories,autonomous driving,augmented and virtual reality pose significant challenges to the bandwidth,delay,and architecture for the wireless mobile communication network.Mobile Edge Computing(MEC)is one of the core technologies to meet the above challenges.It takes advantage of the computing,storage,and communication capabilities of edge network devices to perform distributed processing of services,thereby improving the network resource utilization,and achieving the low-latency and high-reliability trans-mission of services.Computation offloading is a key research issue in MEC.The pros and cons of computation offloading strategies will directly affect the processing power and processing latency to network services.It is the difficulty for designing computation offloading strategies that how to jointly optimize the device selection,resource allocation and network con-figuration to comprehensively consider the energy consumption and latency in different scenarios.This thesis aims at the research on computation offloading strategies in MEC.First of all the research on energy-saving computation of-floading algorithm subject to the latency constraint in MEC is studied.Then,the computation offloading scheme when MEC coexists with non-MEC is stud-ied.After that,the computation offloading mechanism for the coexistence of inhomogeneous mobile applications in MEC is further studied.Finally,this thesis extends the traditional MEC computation offloading scenario where mo-bile applications are independent to the whole interactive scenario where mo-bile applications depend on each other.The main work and contributions are as follows.Firstly,for the high energy consumption of MEC latency-sensitive applica-tions,the more efficient energy-saving computation offloading algorithms sub-ject to the latency constraint in MEC is studied.At first,the optimization prob-lem in the single-device scenario is modeled based on Directed Acyclic Graph(DAG).Then,a greedy task offloading algorithm based on Partial Critical Path(PCP)GA-PCP is proposed,which has the low time complexity.GA-PCP can quickly obtain a suboptimal solution with excellent performance and achieve significant effect on energy saving.After that,the optimization problem in the multi-device scenario is modeled based on the binary offloading model.And the Branch and Bound(BB)method based on Reformulation-Linearization-Technique RLTBB and the greedy heuristic algorithm based on Gini coeffi-cients GCGH are proposed respectively.RLTBB can adjust the accuracy of the solution,but has very high time complexity.While,GCGH has low time complexity.Both algorithms can effectively save the energy consumption of mobile devices.Secondly,for the resource allocation difficulty caused by the significant difference of resource requirement between the MEC and non-MEC,the com-putation offloading scheme when MEC coexists with non-MEC is studied.At first,the optimization problem jointly optimizing the energy consumption and latency of the MEC slice(supporting the MEC computation offloading)and traditional slice(supporting non-MEC mobile applications)is modeled.Then,an information prediction and dynamic programming(DP)based Radio Ac-cess Network(RAN)slicing algorithm IP&DP-RS is proposed.IP&DP-RS can set the uplink/downlink configuration for each slice in advance to adapt the traffic characteristics,allocate resource between and within slices under this configuration,and select devices to offload their computation at the same time.IP&DP-RS can optimize network utility with high fairness and polynomial time complexity,and jointly optimize the energy consumption and latency,thereby improving the performance of computation offloading in MEC.Thirdly,directing at the dynamics of inhomogeneous mobile applications in MEC,the computation offloading mechanism when inhomogeneous mobile applications coexist in MEC is studied.At first,under Time Division Duplex-ing(TDD)mode,the optimization problem jointly optimizing the energy con-sumption and latency for the scenario where the computation offloading and other MEC applications coexist is modeled.Then,for the network configura-tion problem under long time scale,a model-free online dynamic adjustment of TDD configuration C-UCB is proposed.For the device selection and resource allocation problem under short time scale,a greedy resource allocation algo-rithm GRA is proposed,which is embedded in the C-UCB and has low time complexity.C-UCB&GRA can dynamically and efficiently adjust the TDD configuration to adapt the MEC system with inhomogeneous mobile applica-tions.In addition,C-UCB&GRA can properly select mobile devices and allo-cate resources,thereby jointly optimizing the energy consumption and latency.Therefore,the universality of MEC computation offloading is improved.Fourthly,aiming at the interdependence of mobile applications which co-exists in MEC,the computation offloading scheme for the whole interactive scenario in MEC is studied.At first,the factors affecting energy consumption and delay in the whole interactive scenario are analyzed.And the optimization problem for each agent jointly optimizing the energy consumption and latency is modeled.It demonstrates that this scenario has different communication and computing requirements compared with the traditional computation offloading scenario.Then,an online distributed scheme based on Temporal Difference(TD)learning method DMTD is designed,which has the advantages of flexi-bility,speed,and robustness.Each agent can respond to rapid changes of the system in real time by DMTD,thereby achieving efficient performance on com-puting and network.DMTD scheme can make the system cost be converged in simulations,and obtain the excellent long-term performance on jointly opti-mizing energy consumption and delay in an online adaptive manner,thereby expanding the application scenarios of MEC computation offloading.