Research And Application Of Task Scheduling Method In Heterogeneous Video Surveillance Cloud Computing Platform

With the global emphasis on security,and video surveillance data is growing at a massive rate,more and more video surveillance applications are beginning to migrate to the cloud computing platform.In addition,in order to speed up the processing of video tasks and reduce the processing time of tasks,many video processing tasks begin to use GPUs to speed up operations.The current mainstream video surveillance cloud platform based on virtual machine and CPU has the problems of low resource utilization,long task processing time,and small platform throughput.In addition,due to problems such as long startup time and non-programmability of the virtual machine,the monitoring and operation of the platform is difficult and the service quality is poor.This paper studies the video surveillance cloud platform based on CPU-GPU heterogeneous environment.First of all,this paper uses Kubernetes and Docker technology to design and implement a heterogeneous video surveillance cloud platform.Utilizing the lightweight features of Docker,the additional consumption of the platform is reduced,and the resource utilization of the platform is improved.Utilizing the computing and storage resource management capabilities provided by Kubernetes to effectively manage the resources of heterogeneous platforms.In order to facilitate the docking of the video surveillance system and user-friendly access,this paper uses the Golang language cluster management unit module,which exposes the RESTful-style interface while providing a visual operation interface.In addition,in order to further improve the resource utilization of the platform and give full play to the performance of heterogeneous platforms,this paper proposes a two-stage task scheduler CS-NS based on deep reinforcement learning.The scheduler utilizes the advantages of deep reinforcement learning in continuous decision-making problems,combines the task characteristics,data characteristics and resource characteristics of heterogeneous video surveillance cloud platform,learns the optimal scheduling strategy in real time,and selects the appropriate working node and computing device.Finally,experiments show that the heterogeneous video surveillance cloud platform designed in this paper can significantly improve the resource utilization of the platform and increase the throughput of the platform.