Design And Implementation Of Adaptive Deployment System For Edge-cloud Collaborative Model

With the continuous proliferation of mobile smart devices and Io T devices,a large amount of data begins to be generated from the edge side.The cloud computing framework relying on the central server appears to be powerless in the context of big data.The emerging technique Edge computing can meet real-time requirements,but the computing power on the edge side is limited,and many complex model algorithms cannot be run.The edge-cloud collaboration technology that combines the advantages of edge computing and cloud computing has received a lot of attention,especially in the current rapid development of deep learning.The deep learning method of edge-cloud collaboration distributes deep models on edge devices and cloud Servers can give full play to the advantages of edge computing and cloud computing,and balance computing power and real-time requirements.However,in actual production,due to business requirements,the deep model may need to be changed constantly.At the same time,due to network fluctuations,the communication between the edge and cloud may be affected.The deployment of a single edge-cloud collaboration model cannot always meet the production needs.Therefore,this paper proposes a network adaptive edge-cloud collaborative model deployment system,and expounds its related design and implementation.The edge-cloud collaborative model deployment system proposed in this paper mainly realizes three functions,namely model configuration and registration,model selection and collaborative deployment,and model resource monitoring and scheduling.The main function of the model configuration and registration module is to establish a basic model database,including configuring the metadata information of the model and dividing the model into two parts of edge cloud;The model selection and collaborative deployment module is the core module of the system.It is mainly responsible for selecting the optimal collaborative model from the model database according to the actual production environment and user requirements,and deploying the model to the edge and cloud respectively,so as to implement the inference service deployment of edge-cloud collaboration;The model resource monitoring and scheduling module is responsible for monitoring the running of model instances,and scheduling models when resources are insufficient to maximize resource utilization while meeting service requirements.This paper takes the deep face model as an example to show the flexibility of the system for the collaborative model.The experimental results show that the system can greatly improve the deployment efficiency of the edge-cloud model and save the cost of manual maintenance,which makes a huge contribution to edge-cloud collaborative learning.