Research On SDN Resource Optimization Technology For Traffic Engineering

Software-defined networking(SDN)is a new paradigm in networking whose central idea is the separation of control and data plane.With its support on flexible management and agile innovation,SDN can break the rigid architecture of current Internet.Traffic engineering(TE)is a network service which studies measurement,analysis and management of network traffic to improve utilization of network resources,and better meet user's requirements of quality of service.SDN is an ideal architecture to support TE due to its dramatic characteristics,such as logically centralized control plane,open and programmable data plane,and fine-grained flow control.However,such characteristics also confront SDN with severe resource shortage problem when deploying TE application.Recently,there have been several studies on the resource optimization of SDN control plane and data plane,and some encouraging idea such as distributed control plane,elastic control,optimal rule placement and rule set compression have been proposed.However,there are still some challenges when applying these ideas to SDN with TE application,including: 1)the distributed control plane partitions the network in to several domains,thus complicates the flow path control process and multiplies the controller resource consumption,which is called the control stretching problem;2)the rule set generated from fine-grained path control has great redundancies both in forwarding semantics and matching fields,which results in a low resource utilization of the data plane.To address such challenges,this dissertation studies the SDN resource optimization technology for TE,which reduces the resource demand of TE and improves resource utilization through flexible management and distribution of controller resource and efficient compression of rule set.We focus on resource optimization technology of both the control and data plane,to solve the control stretching problem and the rule set redundancy problem respectively.The main contributions of this dissertation are summarized as follows:1)For the control stretching problem caused by network partition of distributed control plane in large scale networks,we propose a flow path aware controller resource optimization mechanism.Firstly,we make a quantitative analysis of the controller resource consumption of a flow request according to the number of control domains that the flow passes through,which concludes that the controller resource consumption can be reduced through adaptively reassigning switches to controllers.Then,we formulate the flow path aware controller assignment problem with the objective of controller resource consumption minimization.Finally,to solve the problem dynamically,we design a sequence assignment algorithm based on the rearrangement inequality.Simulation results show that our mechanism can reduce the controller resource consumption in flow request processing effectively.2)For the control stretching problem caused by network partition of distributed control plane in small scale networks,we propose a horizontal partition based controller load distribution framework.The framework decouples the controller load from the network topology,so that the load can be distributed by directly partitioning the flow space in a fine-grained manner without the assignment of switches,which eliminates the control stretching problem.Furthermore,we formulate the control plane load balancing problem with the objective of minimizing the average response time of the control plane,and design a lightweight online load balancing algorithm.Simulation results show that our framework can reduce the controller resource consumption effectively,and can achieve a nearly optimal load balancing performance in an online manner.3)To address the forwarding semantics redundancy problem due to fine-grained path rule setup,we propose a rule setup mechanism based on path sharing and aggregation.The mechanism decouples the path forwarding semantics from the flow matching semantics and installs the path rules independently,so that they can be shared among all the flows transmitting through this path.Then,it decomposes a path to multiple path segments and aggregates the rules of the paths passing through the same path segment.Besides,it encapsulates the path information into the packet upon its arrival to keep a fine-grained flow control.Furthermore,we formulate the optimal path aggregation problem with the objective of minimizing the number of rules,and solve it with the simulated annealing algorithm.Simulation results show that our mechanism can reduce the number of rules by about 62% with light tag overhead.4)To address the field redundancy problem in fine-grained flow matching rule set,we propose a field cutter based rule set compression mechanism.To adequately exploit the compression performance of our field cutter method,we divide the rule set into reduction set and complete set,and remove the redundant fields in the reduction set to construct a semantically equivalent rule set with fewer fields in some rules.Then,we formulate the optimal field cutter problem with the objective of minimizing the storage demand of the rule set,and develop a bi-directional bound search algorithm with low complexity.Simulation results show that our mechanism can cut down the storage demand of the rule set and reduce the power consumption of rule inserting and flow matching operation.