Performance Management in Software Defined Networking
Examensarbete för masterexamen
Computer science – algorithms, languages and logic (MPALG), MSc
In computer networks, one commonly performs load-balancing assuming fixed capacities of the links in the network. In the microwave backhaul portion of a telecommunications network however, various dynamic effects such as rain and snow affect the capacities of the links. Active measurement of the performance along the available paths in the network can detect these capacity fluctuations. Combining the resulting performance knowledge with an SDN controller makes it possible to reroute traffic around microwave links having reduced capacity, with the goal of balancing the load in the mobile backhaul and improve the overall performance. This thesis simulates a microwave mobile backhaul utilizing SDN and static load-balancing to determine if the addition of active Performance Management can decrease the drop level of the network by enabling dynamic loadbalancing. The proposed design is a dynamic load-balancer which uses an iterative approximation of a congestion minimization algorithm, measuring congestion using endto- end latencies. We evaluate the dynamic load-balancer against a “base case” static load-balancer splitting all traffic evenly across all network paths, and an optimal load-balancer assumed to have exact, real-time knowledge of the link capacities. The results show a clear improvement in drop level as compared to the static load-balancer when simulating a microwave mobile backhaul and modeling the microwave links using an arbitrary link degradation scheme. We show that performing active PM and using the resulting end-to-end latencies as a basis for making dynamic load-balancing decisions can improve the overall performance of a simulated microwave mobile backhaul. The main contribution of this work is the establishment of active PM as a potentially viable candidate for performing dynamic load-balancing in the context of microwave mobile backhauls.
Data- och informationsvetenskap , Computer and Information Science