Converged Networking And Resource Management In Access Networks

The access network is close to users and widely deployed via various technologies.The wireless access technology and the fiber access technology are widely deployed in recent years due to the salient features.Fiber-Wireless(FiWi)access networks converge the wireless access technology and the fiber access technology seamlessly,taking full benefits of high-flexibility,high-bandwidth and low-delay.Besides,due to the development of computation-intensive and delay-sensitive applications,e.g.,Augmented Reality(AR),Virtual Reality(VR),Internet of Things(IoT)and autonomous driving,the computation resources in EC are deployed at the network edge,which are more computationally capable and in close proximity to users,leading to short delay.However,integrating edge computing,the wireless access technology and the fiber access technology is an ongoing research and many challenges remain to be solved for improving access network performance and supporting the applications in access networks.Additionally,to take full benefits of FiWi networks and EC,the resource scheduling algorithm should be well-designed to jointly optimize communication and computation resource allocation for improving resource utilization and application performance.This dissertation focuses on converged networking and resource management in access networks.The main contributions of this dissertation can be summarized as follows:1)The converged networking scheme proposed in this dissertation integrates edge computing and FiWi networks by network virtualization.The converged access networks builed by this scheme can provide both data transmission and computation capability.The computation resources in the converged access networks are deployed at the network edge and FiWi networks enhance the connectivity of multiple edge computing nodes to provide rich flexibility of resource scheduling.The Scale-free Topology Construction algorithm for Low Delay(STCLD)is designed for converged networking and rich flexibility of resource scheduling.Additionally,the Efficient Topology Reconfiguration(ETR)algorithm is proposed to reconfigure the virtual topology suffered unexpected changes.And experiments are performed to validate the STCLD algorithm can construct topologies toward low delay and that the ETR algorithm ensures the network performance of topology reconfiguration.2)The computation resources at a single node in EC are commonly limited,which cannot execute large scale computation tasks.To tackle this problem,an offloading scheme leveraging on neighboring node resources for EC over FiWi networks is proposed in this dissertation.FiWi networks support the offloading transmission with low delay and wide bandwidth.Based on network virtualization,the resource scheduling with the proposed scheme is centralized and flexible.The computation tasks can be offloaded to neighboring nodes.Additionally,a Centralized Resource Scheduling(CRS)algorithm based on the scheme is also proposed to jointly allocate network resources and computation resources.This scheme does not need to deploy new resources,which is cost-effective and quick response.The evaluation results show that the proposed scheme and the CRS algorithm can satisfy more offloading requests and decrease the average completion time.3)The continuous application,e.g.,AR,VR and autonomous driving,involves multiple subsequent user operations and the user continues interacting with the application during the execution,for which the number of user operations,the uploading and downloading data size for offloading computation of each user operation,and the number of central processing unit(CPU)cycles required to execute computation of each user operation are unknown when scheduling resources.The unknown parameters are modeled as random varibles in this dissertation.And an energy consumption constrained average response time minimization problem for continuous applications under uncertainty is formulated.To tackle this problem,the Response Time-Improved Offloading algorithm with Energy Constraint(RTIOEC)is proposed to schedule resources with fewer characteristics of applications.The evaluation results show that the RTIOEC algorithm achieves comparatively short average response time of continuous applications while satisfying the energy consumption constraint with a predefined upper bound of violation probability.The results demonstrate the significant performance of the RTIOEC algorithm in scheduling resources for continuous applications under uncertainty.