Forwarding Strategy And Congestion Control In Named Data Networking

With the advent of the information age,the defects and shortcomings of current Internet architecture have become more and more obvious.However,the incremental improvement solutions could not completely resolve those inherent problems caused by the defective design.Based on the understanding of the development of both network technologies and future network,research communities have tried to build a clean-slate network architecture that can satisfy the needs of future network developments.In this context,Named Data Networking(NDN)has emerged as one of the most competitive solutions of the future network and received more and more attention.So far,the design of the NDN architecture has taken shape,yet researches on the related network tech-nologies still remain to be open.Aim at providing efficient data delivery performance for large-scale content distribution,NDN has introduced many new ideas and functions.However,without taking these new features into account,the existing transport mecha-nisms proposed for traditional network architectures fail to take full advantage of NDN.Therefore,it is of profound theoretical significance and practical value for improving data delivery performance to develop a transmission mechanism that adapts to the char-acteristics of NDN.This dissertation focuses on the problem of data delivery in NDN,and discusses the optimization of network performance from two aspects:dynamic request forward-ing and congestion control,respectively.By being primarily grounded in the principles of stochastic optimization process and Nash bargaining theory,we formulate individual optimization models that take into account the characteristics of NDN for the network problems of interest.Based on these models,practical and feasible solutions are pro-posed to derive the optimal strategies of the corresponding control processes so as to optimize the network performance.Our goal is to provide effective solutions and guide-lines for the problems of NDN data delivery in practice,so as to promote the deployment and application of NDN networks.The main works and innovations of this dissertation are summarized as follows:Firstly,the NDN dynamic request forwarding problem is studied from the per-spective of the overall network and a POMDP(Partially Observable Markov Decision Process)-based request forwarding strategy is proposed using the basic theory of event-based optimization in discrete event dynamic system.We define network status by the number of forwarded unsatisfied requests among network nodes to reflect the usage of network resources,and then specify the events of interest that are related to content requests as well as the feasible actions for each type of event.From these factors,the evolution of the network state is derived,and thereby our POMDP formulations for ND-N request forwarding problem.Our POMDP model can not only make full use of those new features provided by NDN for efficient data delivery,but also be able to effective-ly deal with the inherent uncertainty when making request forwarding decisions in the context of NDN.In addition,a parameterized randomized policy based on observations is adopted as the solution of our POMDP model,which seeks to optimize the long term average expected performance of the network.Given the scale of the problem and the computational complexity of the model,we propose a policy gradient optimization al-gorithm based on the sample trajectory estimation to find the optimal POMDP solution.The simulation results show that the POMDP-based dynamic request forwarding mod-el and the policy optimization algorithm proposed in this dissertation can effectively improve the performance of NDN data delivery.Secondly,since content requests are propagated on a hop-by-hop manner in the context of NDN,we study the request forwarding problem on the data plane of a single NDN router and propose an event-driven optimization method based on the framework of Semi-Markov Decision Process(SMDP).By specifying the definitions of state s-pace,event space and action space,we formulate the problem as a continuous-time semi-Markov decision process with a finite state space.Our SMDP abstract is able to achieve load balance among all interfaces and maximize the utilization of link resources by allocating the limited network resources in a more reasonable fashion,since it takes into consideration the state of all interfaces when making forwarding decisions and pro-vides differentiated services for different types of content requests.For the solution of our SMDP model,we propose a neural networks based Q-learning algorithm,which takes advantage of the adaptation ability of Q-learning algorithm and the function ap-proximation ability of neural networks to deal with the problems of "model disaster"and "dimension disaster" that traditional theoretical calculation methods and tabular reinforcement learning methods may encounter.The effectivenesses of our proposed SMDP formulation and the corresponding algorithm are evaluated from many aspects through simulations.The simulation results show that our SMDP-based approach can effectively deal with the characteristics of NDN,while achieving efficient adaptive re-quest forwarding in comparison with existing forwarding strategies.Lastly,based on the cooperative game theory,we cast the congestion control prob-lem of NDN within a Nash bargaining framework.Since NDN provides a strong hop-by-hop flow balance,it cannot afford to neglect the influences of both Interest packets and data packets on network congestion due to their interdependence.For this reason,we establish a bargaining model wherein the Interest and data flows between a pair of routers compete for the limited link bandwidth with the objective of maximizing the Nash product of individual utilities of flows.With the method of Lagrange dual de-composition,the optimization problem is decomposed into independent subproblems that could be solved in parallel on both side of the router pair,and thereby the dual opti-mization problem can be also derived,if the utility functions of flows are well designed.Due to the independence between variables,the subproblems are further split into a set of flow rate optimization problems through the local Lagrangian method.Finally,on the usage of Nash bargaining solution for the dual optimization problem,we derived a hop-by-hop flow rate control scheme based on Nash bargaining solution.Afterwards,we have a discussion on the related practical issues which serves as a guideline on how to implement our congestion control scheme.The simulation results show that the pro-posed congestion control scheme not only make full use of the link resources,but also acheives proportional fair sharing of link resources among competing flows.