Delay-Aware Resource Allocation For Partial Computation Offloading In Mobile Edge Cloud Computing

In recent years,advancements in cloud computing,edge computing,Internet of Things(Io T),and fifth generation communication technology(5G)have made it possible for Io T end devices to efficiently run computation-intensive and delay-critical tasks or provide massive data collection services.Mobile Edge Cloud Computing(MECC),a very promising paradigm for partial computation offloading,not only takes advantage of the low latency of edge computing,but also provides powerful computing capabilities and huge storage space of cloud platforms.Therefore,end devices are able to offload some complex computing tasks to edge or cloud servers,or process the huge amount of collected data on them.However,a challenging delay-sensitive optimization problem of two-stage tandem queues will inevitably be encountered when designing resource allocation strategies for MECC.Specifically,due to the coupled resource allocation of edge and cloud processing queues,it is difficult to guarantee the end-to-end latency of computing tasks of multiple end devices with minimal resources.In this thesis,we investigate the problem in terms of the stochasticity of computation request arrivals,the stochasticity of service time,and the dynamics of computation resources.We first model the MECC network as a two-stage tandem queue system consisting of two sequential computation processing queues with multiple servers.A Deep Reinforcement Learning algorithm is then applied to learn a computation speed adjusting policy for the tandem queue system,which provide end-to-end delay insurance for multiple offloading tasks while preventing excessive use of computation resources of edge and cloud servers.In addition,due to the coupling of resource allocation between tandem queues,we incorporated Gated Recurrent Unit(GRU)to the RL network structures,which captured the sequential relations between two queues so that the goal of improving performance is achieved.Traditional healthcare system usually has difficulty in providing low-latency vital sign monitoring services for multiple patients at the same time,cannot process large amounts of data in a timely manner,and cannot provide intelligent diagnosis services.Therefore,we apply the proposed approach to real-time medical services provisioning in the healthcare cloud,which effectively reduces the waste of computation and communication resources through a reasonable resource allocation scheme.It not only improves the patient’s medical experience,but also provides help for doctors to study diseases and formulate diagnosis and treatment plans.Finally,we conduct extensive simulation experiments,and the results show that our method can reduce the use of resources by 15% on average.Our approach is able to satisfy the latency requirements of offloading tasks while avoiding overuse of resources even in dynamic network environment.