Controller placement methods analysis

摘要

Software-Defined Networking (SDN) offers flexibility and programmability to the network infrastructure through the introduction of a controller. However, the controller introduces extra delay into the system as new data flows must query the controller for instructions of how to route traffic. This becomes an increasing problem for large scale and delay sensitive networks such as those found in high-criticality infrastructure. The delay introduced can be minimised by optimal placement of the controller or decreased further by introducing additional controllers. Although the problem of optimal placement for multiple controllers is known to be NP hard, approximations can be used. The analysis of four different methods has therefore been conducted and looks at the scalability, through the lens of complexity. It is found the four methods, full search, linear programming, local search and an adapted version of the k-means++ algorithm, vary significantly in their complexity. It is also found that the accuracy of the methods varies with the complexity, creating a definitive trade-off between the two attributes.