Research On The Reinforcement Learning-based Algorithms For Deploying SFCs In Dynamic Networks

Network Functions Virtualization(NFV)technology is one way to address the increasing difficulty of providing and managing Network services.NFV separates network functions from hardware and deploits them as software on a common server.This technique can be used to improve service flexibility and secure user data.Therefore,many aspects of NFV technology have been extensively studied.This thesis mainly studies two problems: Service Function Chaining(SFC)deployment with load balancing in dynamic single-domain network and dynamic SFC deployment with energy saving in dynamic multi-domain network.For the problem of dynamic SFC deployment under dynamic single-domain network with load balancing,this thesis proposes a q-learning single-domain dynamic SFC deployment algorithm,which combines with q-learning in reinforcement learning to solve the problem.Problem can be decomposed into two steps,the first step,the algorithm of reinforcement learning module is responsible for the alternate path set training and output,some meet the requirements of the deployment path was generated according to user requirements,the second step,load balance module is responsible for grading alternate path,investigate its influence on the network load balancing,and selected the optimal solution from alternative paths in the collection.In the process of solving the problem,this thesis also modified the model building and the training algorithm of Q learning for the problem,greatly improving the training efficiency to adapt to the change of dynamic network.The results of the following simulation experiments show that the algorithm can output the approximate optimal solution in a short period of time,achieve the goal of maximizing the benefits of the service provider,and consider the load balance of the network.For the problem of dynamic SFC deployment in dynamic multi-domain network,which gives consideration to network energy saving,this thesis proposes Q learning multi-domain dynamic SFC deployment algorithm,and upgrades the results of the previous algorithm.Multi-domain network layer in the first place,to form the top abstract network and each subdomain networks,then respectively to train the network using Q learning,when the user service requirements comes,in the upper and lower two layers of network output by alternative paths,and then consider the subdomain privacy protection,the calculated fuzzy value is uploaded after each subdomain is selected,finally got every fuzzy values of the path network energy saving module,computing the total score,output the best solutions.After the simulation experiment,it can be concluded that under the condition of successfully protecting the subdomain privacy,the algorithm can output the SFC deployment scheme with low average energy consumption in a short time,and the algorithm has a high success rate of deployment,thus ensuring the high profit of the service provider.