Research On Containerized Deployment Management Of Heterogeneous Resources On Embedded Platform

Higher and higher requirements have been put for the computing capabilities of edge devices by the Internet of Things and edge computing.The appearance of embedded heterogeneous multi-core systems connected by an on-chip AXI bus or an inter-chip PCIe bus can better meet the system's demand for high computing power.However,high system consumption and poor flexibility are regarded as problems of existing customized methods for device ac-cess and management.At the same time,most of the current deployment strategies devote to improving the quality of system services,and the limited computing resources of the em-bedded platform and the heterogeneous computing nodes have been ignored.Resulting in low system resource utilization and limiting the improvement of system performance.In or-der to optimize the registration management of embedded heterogeneous resources and the placement of application services,the device plugin framework and microservice software development model have been used by this thesis.The specific work is as follows:(1)In view of the registration and management of heterogeneous resources problems on the edge embedded platform,lightweight device plugins based on the hardware resources of the on-chip AXI bus and the inter-chip PCIe bus have been designed respectively.Device plug-ins are optimized based on the Kubernetes plugin framework and communicate with a cluster in a client/server manner.A connection is established and a request is sent in the registration phase by the device plugin,which acts as a client.A request is received and a resource is allocated in the management phase by the device plugin,which acts as a server.To illustrate the impact of the containerized device plugin on the performance of the embedded platform,some performance tests on the embedded platform have been executed under different con-ditions.The results show that the performance only drops by 1% for CPU and memory,and the network throughput drops no more than 2% under certain conditions in a containerized environment.(2)To face the challenges brought by heterogeneous computing systems to service place-ment,a service placement strategy based on the tendency of heterogeneous resource requests is proposed.Firstly,the application is divided into the form of microservices,and DAG is used to describe the communication between services.Secondly,the request tendency of different resources is determined by the ratio of the resource size requested by the service to the total resource size contained in the computing node.Through normalizing the above ratio,the request propensity weight is calculated.Finally,based on the above weights,a weighted summation method is used to calculate the system's heterogeneous resource uti-lization.Through experimental comparisons with different application numbers,network layers,and routing nodes,the results show that compared with the strategy based on service request rate,the strategy proposed in this thesis improves the utilization of system hetero-geneous resources by nearly 10%,and the network bandwidth usage has dropped by nearly25%.On the one hand,the research of this thesis enriches the hardware resources supported by the existing container orchestration environment and provides a feasible solution for the reg-istration and management of embedded resources.On the other hand,to a certain extent,it provides a new solution to the problem of service placement in heterogeneous comput-ing systems.The research in this thesis provides a reference for the future deployment and management of software and hardware resources in a containerized manner on an embedded platform and is a useful exploration of the application of container technology in an embed-ded environment.