Research On Collaborative Resource Scheduling Mechanism In Computing Power Network

The Internet of Everything is considered to be an important development direction of the next generation of the Internet.In order to meet the huge computing power demand for information processing by future intelligent applications,Computing Power Network(CPN)has been proposed as a promising computing paradigm.Based on ubiquitous network connectivity,CPN allocates idle computing,storage,network,and other resources in the system through the control plane to provide computing power resources required by upper-layer services,which has great potential to realize on-demand task scheduling and flexible resource sharing.However,as the network scale continues to expand and the complexity of the network continues to increase,large-scale collaboration among multiple computing power service nodes across operators or regions has become extremely challenging due to workload imbalance and limited resources of single service nodes,etc.How to efficiently control and utilize these distributed ubiquitous computing power resources is an urgent practical problem to be solved.Therefore,this thesis takes resource collaboration in the CPN-enabled scenario as the background and specifically studies the task scheduling mechanism to realize multi-node collaboration,with a view to improving the resource utilization of the computing system while ensuring the effectiveness of system task processing.The specific work is as follows.(1)Taking into account the deployment of relevant functional entities in the computing system,task arrival situation,and task processing status of service nodes,the multi-node collaborative,the integrated system architecture of computing network under two different system working modes,"scheduling first and queuing later" and "queuing first and scheduling later," are discussed and designed separately to achieve the interoperability and integration of computing and network capabilities.(2)Based on the proposed CPN-enabled multi-node collaborative,integrated system architecture,this thesis further studies the collaborative task scheduling mechanism between different computing nodes.What’s more,from the service provider’s perspective,this thesis aims to comprehensively optimize the task processing delay and load balancing performance.(3)Taking the system architecture of "scheduling first and queuing later" mode as a reference,this thesis first designs a dynamic task scheduling scheme based on CPN.Considering the dynamic and continuous changes of the system,the task scheduling problem is described as a Markov Decision Process,and a Deep Reinforcement Learning-based task scheduling algorithm is further proposed to solve the problem.(4)A priority-aware task scheduling scheme is further designed based on the system architecture of the "queuing first and scheduling later"mode and inspired by the idea of "Sliding Window." Evaluating the scheduling priorities of different tasks in the system before scheduling and then making scheduling decisions based on the priority of each task can further improve the task processing success rate of the computing system.In summary,this thesis focuses on the collaborative scheduling mechanism in CPN.Firstly,the designs of CPN-enabled multi-node collaborative integration architecture are carried out,and then applicable task scheduling schemes are designed based on different system working modes.Finally,the simulation results verify the effectiveness and superiority of the proposed scheme.