Research On Service Migration Framework And Methods For Edge Computing Resource-constrained

Edge computing paradigm(ECP)has become one of the most innovative modes of processing distributed the Internet of Things(Io T)data after the cloud computing paradigm(CCP).In ECP,edge computing devices are located between the Io T device and the cloud center,and closer to the Io T device.Offloading services from the cloud center to edge computing devices reduces service latency and data traffic across the Internet.However,as a service runs,its demand for resources changes dynamically.When the demand for resources of the service run by the edge computing device is greater than the resources owned by the edge computing device,the edge computing device is called resourceconstrained.Resource-constrained edge computing devices cannot meet the resource requirements of the service,which will reduce the Quality of Service(Qo S)offloaded to the edge computing devices.In view of the limited resources of edge computing device,this paper focuses on the research of service migration framework and method for edge computing resource-constrained.The main research contents focus on the following two aspects:(1)A resource-constrained service migration framework for edge computing is constructed.The basic idea is: when the resources of edge computing devices are limited,the services can be migrated to edge computing devices with abundant resources to ensure the normal operation of the services and avoid Qo S decline.In this paper,from the perspective of service migration,a resource-constrained service migration framework for edge computing is constructed.Based on this framework,edge dynamic service migration algorithm dynamically migrates services to resource-rich edge computing devices by evaluating the resources of edge computing devices and the consumption of edge services on resources.In order to assist the edge dynamic service migration algorithm,four auxiliary models are proposed: edge computing device resource consumption model,edge service resource consumption model,edge computing device selection model and edge computing device selection model.Finally,based on the framework and the corresponding edge dynamic service migration algorithm,an edge service migration prototype system is implemented,and a service migration case experiment with 10 edge computing devices is presented on the prototype system.Experimental results show that the framework not only increases the running time of edge services on edge computing device,but also reduces the service delay,improves the resource utilization of edge computing device,and avoids the decline of Qo S.(2)A lightweight action recognition method based on depth-separable convolution is proposed.The resource limitation of the edge computing device is mainly caused by the fact that the service managed by the edge computing device requires more resources than the resource owned by the edge computing device.Therefore,in order to avoid the limitation of edge computing device resources,a lightweight action recognition method based on depthseparable convolution is proposed from the perspective of reducing the demand of services on resources for specific services such as video action recognition.The method reduces the need for storage resources by removing redundant video frames.The computational complexity of the video action recognition model is reduced to reduce the need for computing resources.The experimental results show that the method can reduce the demand of storage resources and computing resources by about 30% while ensuring the accuracy of action recognition.Finally,based on this method,a prototype system of action recognition is implemented,and it is called as smart camera.