Research On Management And Deployment Of Computing Resource Based On Software Defined Architecture

This paper focuses on the management and deployment of computing resource based on Software-Defined Architecture.The quick development and popularization of heterogeneous systems and heterogeneous chip technology are the reason for the modular orientation of the new software defined architectures.Many management problems need to be solved in multi-channel communication platforms based on the modular architecture,such as the computing resource competition,the load unbalances,the optimization requirements caused by the deployment diversification.Therefore,it is of great significance to study a software defined modular architecture system and optimize the management and deployment of computing resource based on the architectureThis paper deeply analyzes the development of software defined architecture at home and abroad and explores the characteristics of software defined architecture and reasons why they become modular.The modular architecture will be used widely in the heterogeneous communication systems in the future.So,it is necessary to design a software defined modular architecture and manage the modular resource.Then,this paper designs a Software-Defined Modular Architecture and proposes a reconfiguration deployment method to flexibly control and manage the development of heterogeneous systems.The effectiveness of the proposed model is verified on heterogeneous systems.Finally,aiming at the resource imbalance and competition caused by the heterogeneous system based on the modular architecture,this paper designs and implements a GPP resource deployment strategy based on Reinforcement Learning to optimally deploy GPP resources.The experiment results show that the system performance and scalability of the communication system are improved.The main work and innovations of this paper are as follows:(1)This paper studies and analyzes the characteristics and reasons of modular development of software defined architecture.This paper deeply studies a variety of typical software defined architectures of the U.S.military and other institutions.After comparing and analyzing the specific modular design schemes of different software defined architectures,it summarizes that the idea of modulation has gradually expanded from the application layer to the whole architecture,even to hardware modules.The application direction and scenario of the architecture is further explored.The reason for modulation is the cooperation of large-scale heterogeneous systems and the popularization of heterogeneous chip technology.So,it is necessary to design a software defined modular architecture and manage the modular resource.(2)This paper proposes a Software-defined Modular Architecture and a related reconfigurable deployment method.According to the current situation and requirements of the Software Defined Radio(SDR)platform,a Software-defined Modular Architecture based on both SCA and the local advantage idea of some related architectures is put forward.The overall framework of Software-defined Modular Architecture(SDMA)is proposed.The key methods of each SDMA layer and the related reconfigurable deployment mode are introduced,as well as the construction and connection of SDMA modules.The new architecture is carried out in heterogeneous systems to transplant and reuse applications across platforms.(3)This paper proposes a GPP Resource Deployment Strategy based on Reinforcement Learning.This paper describes the requirements of modular resource management with the architecture above,then designs and implements a resource deployment algorithm based on Reinforcement Learning for GPP resources.Firstly,according to the resource attributes,the GPP device resources and components in the platform are modeled as a matrix,including matching resources and allocating resources.Then,a Reinforcement Learning environment is constructed to realize the iterative adaptation of GPP devices and components.Finally,the optimized resource deployment strategy is got.Compared with the traditional deployment scheme,the proposed algorithm improves the system performance and resource utilization.At the same time,the best deployment state of the system could be got through the comparative experiment of component deployment.