Study On Congestion Controls In Software Defined Network

Software Defined Network (SDN) provides the possibility of offering more adaptive and effective congestion control with the help of its centralized control ability and global perspective. Compared to end-to-end schemes, Active Queue Management (AQM) works in the forwarding plane and is more suitable for performing global congestion optimization with the help of SDN, which also focuses on the forwarding plane. This thesis proposes an AQM adaption framework in SDN called OpenAQM. OpenAQM uses the global perspective of SDN to optimize AQM performance, including performing dynamic AQM parameters adjustment and AQM algorithm switching.AQM parameter adjustment needs certain guildines. In this thesis, we use classical control theory method to build the model of TCP/AQM system and use Nyquist criterion to analyze the stability condition of the system and thus perform parameter adjustment. We describe the common procedure for building the stability condition and analyze RED algorithm to gain specific rules to adjust RED’s parameter. We perform NS2simulation to give proof of our analysis.Switching between AQMs needs certain evaluation for AQM algorithms. By refering to the duality model of TCP/AQM, we can obtain the equilbrium performance and use that as evaluation metrics to perform algorithm switching. We describe the common procedure to obtain the equilibrium performance and give detailed evaluation method in a single bottle-neck network and perform NS2simulation to validate our evaluation of RED, PI and REM. Further, we perform dynamic algorithm switching by taking advantage of the mentioned algorithms. Simulation results show major improvement in responsiveness is achieved in our scenario.The main innovation of this thesis is to propose an AQM adaptation framework in SDN called OpenAQM. Compared to other congestion control schemes in SDN, we use the approach of queue management to solve congestion problem in SDN context. It utilize suitability of AQM and SDN. It not only lowers down the difficulty of system deployment, but also offers the capability of flexibly control AQM to solve congestion problem. On the other hand, based on classical control and optimization theory, this thesis gives specific guildlines for AQM optimization. We analyze RED, PI and REM in practice and perform NS2simulations to show the feasibility and effectiveness of OpenAQM.