Performance evaluation of flowtable eviction mechanisms for software defined networks considering traffic flows variabilities

Abstract

Software Defined Networks (SDN) is a new paradigm that emerged to improve network management through separation of control from the data plane using a standardized protocol. OpenFlow is the most popular standard to achieve the benefit of SDN. For every arrived flow, a corresponding flow entry is installed in the switch flowtable to guide the data transmission process. The proliferation of the Internet of Things (IoT) devices increases the number of flows generation unfortunately switch flowtable is constraint with limited space. Consequently, it led to flowtable overflow. Several studies leverage the First in First Out, Random replacement policy to removed old flow entries when the flowtable is overflowed. Although some performance gains were reported. However. Flows exhibit variabilities in terms of duration, inter-arrival-time, and the number of packets differs. Usually, some flows contain a large number of packets others have few packets. Applying these replacement policies may not always meet the demand of these types of flows. As such, this study experiments eviction mechanism to evaluate the performance of the two schemes and observe their eviction behavior with respect to flow features. On average FIFO preserved flows with a large number of packets by 4.95% while Random preserved by 36.11%. In conclusion, Random showed better performance compared to FIFO with respect to an increasing flow generation and change in flowtable size.