Research On Resource Adaption Mechanisms For Dynamic Service Function Chaining Based On Smart Identifier Network

With the rapid development of the Internet,network application scenarios are increasingly diverse,where users have different demands on network services provisioning,including ordered service functions such as firewall,intrusion detection system and performance enhancement proxy,and quality of service such as bandwidth and delay.In traditional networks,service functions are usually coupled with dedicated hardware and network topologies,which have drawbacks in inflexibility and high cost of deployment.Therefore,with the rise of emerging network architectures and technologies,the research on dynamic and on-demand service function chaining has attracted much attention in recent years.However,problems and challenges in this field have not been completely solved,where related mechanisms and technical details still need to be improved and refined.Based on the architecture of ‘Smart Identifier Network',this dissertation conducts an intensive study on the dynamic service function chaining and its resource adaptation mechanism.The main contributions of this dissertation are three-fold shown as follows.(1)To realize dynamic and on-demand provision of service functions in multidomain networks,we propose a context-aware service functions chaining mechanism,which enables fine-grained traffic awareness and dynamic service strategy.Specifically,based on the Network Service Header(NSH)protocol,we first design a context header allocation scheme for hierarchical multi-domain networks,which stores and shares user locations,application types,attack behaviors and other relevant information.Besides,different traffic steering mechanisms are designed for dynamic context-based service combination in intra-domain and inter-domain scenarios.Secondly,a heuristic mapping algorithm is proposed to save bandwidth resources for context-aware and dynamic service function chains.Experiments and simulations demonstrate that the proposed chaining scheme can realize flexible and fine-grained services,and the mapping algorithm can effectively improve the cost-effectiveness compared with node-load-balance greedy and tabu search algorithms.(2)To alleviate imbalance of resource consumption on nodes and links in edge networks,where user requests have different priorities and various resource demands,we propose an SFC mapping strategy combination scheme.We first formulate the mapping of multiple service function chains with priorities as a multi-step Linear Integer Programming(ILP)problem,of which the mapping strategy(i.e.,the objective function)in each step is configurable to improve overall CPU and bandwidth resource utilization rates.Secondly,to solve the strategy selection problem in each step and alleviate the complexity of ILP,we propose an adaptive deep Q-learning based mapping approach,where an agent is learned to make decisions from two low-complexity heuristic mapping algorithms.Finally,we conduct extensive simulations and related results demonstrate that compared with a single strategy or random selections of strategies under the ILP-based approach or the proposed heuristic algorithms,our approach accepts more requests and improve whole-system resource efficiency after limited training episodes.(3)To solve the problem that it is infeasible to deploy service function chains in overlay networks using optimization methods without knowledge of underlying network configurations(e.g.,topology and routing configurations),we propose a policy gradient based approach for service path selection selections in overlay networks.Specifically,we first formulate the problem into an ILP model with the objective of maximizing revenue,where underlying network configurations and multiple service function chains are given in advance for benchmarking.Secondly,we present the online service function chain path selection into a Markova decision process and propose a corresponding policy gradient based solution.Numeral simulations demonstrate that our approach has better performance than greedy,random selection,and deep Q-learning based approaches,which deploys more requests and obtain more revenues before resource competition happens,close to the optimal result of ILP-based method.Besides,we build a simulation platform and verify the feasibility and efficiency of our approach.