Cooperative Utilization Of Dynamic Bandwidth And Computing Resources Within The Network

With the development of edge computing,bandwidth and computing resources(B&C resources for short)are continuously merging.Network transmission infrastructure is shifting from solely providing connection services to providing both connection and computation services.The collaborative use of in-network B&C resources to achieve computing while transmitting has attracted widespread attention,including the fields of in-network computing(COIN),which focuses on computation within network links,and compute-first networking(CFN),which prioritize wide area B&C resources.However,the scope of B&C resources is constantly expanding,and their geographical distribution and dynamic characteristics are becoming increasingly prominent.Effective matching of B&C resources has become a key factor in their collaborative utilization.Current research on matching and collaborative use of B&C resources still faces the following problems:(1)The evaluation of resource matching is a prerequisite for resource selection and task computation.When B&C resources change dynamically,it is worth considering whether they still meet the task’s resource requirements.Current research mostly considers changes in bandwidth resources,and further expansion is needed to consider changes in computation resources in the same time.(2)In in-network computing,fixed computing nodes are often used.However,there are performance differences among nodes,and further research is needed to match the optimal computing nodes in a distributed,flexible,and accurate manner.(3)When resource utilization expands from in-network to wide area network scope,effective range constraints on resources become the key to avoiding global perception and scheduling,and improving matching efficiency.Current research mostly adopts fixed range constraints,but further attention should be paid to partitioning the effective range and determining the update interval for resource status due to insufficient dynamicity of B&C resources.(4)In addition,with the new network developing,relevant open-source simulation platforms are lacking.More attention has been paid to the completeness of network protocol stack or the coverage of computation scenarios,with relatively little attention given to the integration of both network and computation.In response to these problems,this article focuses on the evaluation of B&C resource matching in collaborative computing,and distributed matching of B&C resources in in-network and wide-area networks.Two extended simulation platforms represented for different network architectures for collaborative use of B&C resources has been designed,and the main research contents and innovations are as follows:(1)Regarding the problem of whether B&C resources still meet the task requirements when they change dynamically,a stable matching evaluation model is proposed by considering the characteristics of non-conservation of data flow introduced by B&C resources,and task computation.First,the delay threshold constraint for task computation is transformed into a resource state solution set that meets the demand for B&C resources.Next,the optimization process of the lower bound of the state solution set is explored.Finally,by using the discrete probability distribution of resource state to represent the dynamic nature of B&C resources,the joint probability that satisfies the lower bound is solved to obtain stable matching degree.The evaluation process examples and application in edge collaborative reasoning demonstrate that this model can reduce the delay jitter and task failure caused by resource dynamic changes during task computation and provide a basis for selecting B&C resources.(2)Regarding the inflexibility and inaccuracy of distributed matching of computing nodes in in-network computing,a two-stage computing resource matching and optimization process is designed and implemented.In the first stage,dependency information is obtained using information-centric networking and piggybacking technology.Then,the optimal node for task execution is accurately selected by evaluating,comparing,and updating in a hop-by-hop manner.In the second stage,the result is directly calculated and returned through the selected node.Furthermore,to avoid traversing all nodes in the network,a dynamic pruning strategy for B&C resource traversal space is proposed by considering factors such as resource capacity,task size,and data transmission size.Experimental results show that this method can achieve efficient and accurate distributed matching of computing nodes in in-network computing.(3)To address the effective range constraint and resource status update interval issues when matching B&C resources in a wide area network,a distributed B&C resource sensing and collaborative utilization framework is proposed.Starting from tasks with time delay constraints,a distributed measure of service capabilities representing B&C resource status information is designed,and the effective coverage range of B&C resources is dynamically identified based on time delay requirements.This framework distributively achieves both instantaneous and stable resource matching indicators and stop diffusion conditions of status information in the network.For task requests,the pruning effect of traversing B&C resources is also utilized.A more general analysis process for the resource status update window size is proposed,and the relationship between upper bound of window size and service success probability is given.Finally,the experimental results demonstrate the effectiveness of the proposed framework and methods,ensuring the effectiveness of status information and avoiding global flooding of status information.Additionally,the stability matching of the resource utilized in this thesis is higher than the instantaneous matching in terms of task success rate,implying the latter is a special case of the former.(4)Focusing on the relative isolation in the utilization of B&C resources in the current simulation platforms.For information-centric network,based on the information-centric network simulator ndnSIM,we achieve application-layer computing logic and scalable custom flow-level forwarding strategies without affecting the existing data forwarding process to enable processing of data at intermediate computing-capable forwarding nodes.For TCP/IP network,to flexibly combine heterogeneous devices and meet the need of collaborative computing,the computing layer is abstracted into different functional modules and intermediate queues for caching tasks based on the network simulation platform NS-3.On this basis,different distributed computing architectures can be further constructed.The functional validation and application cases support the ability of the two simulation platforms proposed in this thesis to collaboratively utilize B&C resources.The related code is open for use,which promotes the development of research ideas related to edge computing and in-network computing.