Early flow table eviction impact on delay and throughput in software-defined networks

Abstract

In a software-defined network (SDN), the forwarding rules are installed at switch's flow tables that are built using Ternary Content Addressable Memory (TCAM). TCAM has a limited storage capacity which causes flow tables to overflow which degrades the performance of SDN. Early eviction can reduce the flow table overflow problem to maximize their usage. This paper analyzes the impact of early eviction of flow entries from the flow tables before overflow happens. The analysis is based on three schemes, namely First-In First-Out (FIFO), Random, and Least Recently Used (LRU). These schemes are used for early eviction at certain threshold values. The values decide the flow table capacity at which the flow entries start to evict before it reaches overflow. We have used the flow tables of limited size and Distributed Internet Traffic Generator (D-ITG) to inject the traffic. Our results show the decrease in delay and increase in throughput in the case of early eviction of flows from flow tables as compared to the normal eviction. In all schemes, the LRU shows the best results to minimize the delay that occurs due to the extensive communication between switch and controller.