Search On Parallel Network Service Function Chain For URLLC Applications

The forthcoming fifth-generation wireless communication system(5G)is expected to support a wide range of emerging applications over cellular mobile communications.One of its key use cases is Ultra Reliable and Low Latency Communication(URLLC).A common view among active researchers from both academia and industry is that URLLC will play an important role in providing new services and applications for vertical areas such as factory automation,autonomous driving,etc.The key performance metrics related to URLLC include latency,reliability,and availability.Obviously,achieving such strict delay and reliability requirements is extremely challenging.In addition to the evolutions of air interface technology,novel network architecture needs to be devoloped.Software Defined Networking(SDN)and Network Function Virtualization(NFV)are important solutions for the core network architecture,so that the network functions are not dependent on proprietary hardware,and are committed to providing flexible and diverse service requirements.However,NFV may compromise certain performance to some extend,and latency may be increased with network functions,which is intolerable for delay-sensitive services.The proposed Network Function Parallelism(NFP)architecture makes it possible in effectively reducing end-to-end delay.Therefore,it is imperative to study the deployment and scheduling of service function chain to reduce latency while maintain high reliability based on parallel network functions.First,the Parallel Network Service Function Chain deployment mechanism is studied.In the computing and bandwidth resource-constrained networks,the reasonable deployment of the Parallel Network Service Function Chain can reduce the end-to-end delay.Aiming at minimizing the processing delay and link delay,this thesis establishes a mathematical optimization model,and describe the heuristic solution based on genetic simulated annealing algorithm.The simulation comparison shows that the introduction of the NFP architecture reduces the end-to-end delay of the service by more than 24%,and the deployment scheme obtains a higher probability of service request success,but brings consumption of bandwidth resource.Based on the deployment,next consider the reliability of the Parallel Network Service Function Chain.High reliability is another important indicator of URLLC,so considering the reliability of services is also a critical issue.Due to the unreliability of physical nodes and virtual machine programs,the entire service request is unreliable.This thesis uses a dedicated backup mechanism to provide redundant network function backup nodes to improve end-to-end reliability.Aiming at minimizing the reservation of computing resources and bandwidth resources,this thesis establishes an Integer Programming model with the constraint of URLLC reliability index.Considering the particularity of the problem,this thesis convert the problem to a Markov decision problem,and a Q-Learning enhanced learning algorithm is proposed.The simulation proves that Q-Learning can achieve a reliability level at 99.999% with the lowest resource consumption,when considering the increasing in latency caused by backup switching.Finally,considering that most URLLC services are provided through Mobile Edge Computing(MEC)networks,the low-latency scheduling problem in MEC networks is studied.Due to the scarcity of MEC network resources,a reasonable scheduling Parallel Network Service Function Chain can maximize the utilization of computing resources and minimize service completion delay.A lot of attention has been paid to the problem of function deployment and scheduling in NFV.Nevertheless,existing research only considers the priority constraints between VNFs in the service.Research on the interactions between services is still lacking.Considering that Network Functions can be executed in parallel,an efficient deployment and preemptive scheduling strategy for the Parallel Network Service Function Chain is proposed.In this thesis,the problem is modeled as an Integer Programming problem,and the Branch-and-Bound algorithm is used to solve the optimal solution.To reduce the complexity of the problem,a VNS heuristic scheduling algorithm is proposed.The simulation proves that the algorithm obtains a solution close to the optimal solution,and is more suitable for mapping and scheduling of multiple Parallel Network Service Function Chains.Both network function parallelism and preemptive mapping scheduling schemes can reduce latency.