[Paper Review] Performance analysis of SDN controllers: POX, Floodlight and Opendaylight

Gereltsetseg Altangerel, Tugsjargal Chuluuntsetseg+1 more|arXiv (Cornell University)|Dec 20, 2021

Software-Defined Networks and 5G4 citations

TL;DR

This paper evaluates the performance of three open-source SDN controllers—POX, Floodlight, and OpenDaylight—using MINinet to emulate mesh and tree network topologies. It measures TCP and UDP throughput and first-packet RTT, finding that OpenDaylight outperforms POX and Floodlight in throughput and latency, particularly in complex topologies.

ABSTRACT

The IP network is time-consuming for configuration and troubleshooting because it requires access to every device command line interface (CLI) or Graphical User Interface (GUI). In other words, the control plane (gathering information to forward data) and data plane (data forwarding plane) are run on every intermediary device. To solve this problem, the software defined network emerged and separated the control plane (done by software) from the data plane (done by hardware) [1]. In addition, the control plane is operated in the central point named the controller. There are many controller software programs in simulation and production network environments. In this paper, we compare three open source controllers called POX, Floodlight and Opendaylight (ODL) in simulation network created by MININET SDN emulation program in terms of TCP and UDP throughput and average Round Trip Time (RTT) of the first packet of the flow in the mesh and tree topology.

Motivation & Objective

• To evaluate and compare the performance of three widely used open-source SDN controllers: POX, Floodlight, and OpenDaylight.
• To analyze how controller choice affects network performance in different topologies, specifically mesh and tree structures.
• To measure key performance indicators such as TCP and UDP throughput and average Round Trip Time (RTT) of the first packet in a flow.
• To provide empirical insights into controller scalability and responsiveness under varying network conditions.
• To support network architects in selecting optimal SDN controllers based on real-world performance metrics in emulated environments.

Proposed method

• The study uses MININET to emulate software-defined networks with controlled configurations of mesh and tree topologies.
• Three open-source SDN controllers—POX, Floodlight, and OpenDaylight—were deployed and evaluated under identical network conditions.
• Network performance was measured using TCP and UDP traffic flows to assess throughput and first-packet RTT.
• The experiments were conducted across multiple runs to ensure statistical reliability and consistency of results.
• Performance metrics were collected and compared across controllers and topologies to identify relative strengths and weaknesses.
• The evaluation focused on controller responsiveness and data plane efficiency under varying load and topology complexity.

Experimental results

Research questions

• RQ1 How do POX, Floodlight, and OpenDaylight compare in terms of TCP throughput across mesh and tree network topologies?
• RQ2 What is the UDP throughput performance of each controller in the same emulated network environments?
• RQ3 How does the average Round Trip Time (RTT) of the first packet in a flow vary across the three controllers in different topologies?
• RQ4 Which controller demonstrates the best scalability and responsiveness in complex network topologies?
• RQ5 Does topology structure (mesh vs. tree) significantly impact the performance differences among the three controllers?

Key findings

• OpenDaylight achieved the highest TCP throughput in both mesh and tree topologies, outperforming POX and Floodlight.
• In UDP throughput tests, OpenDaylight also demonstrated superior performance, particularly in high-load scenarios.
• The average RTT for the first packet was lowest in OpenDaylight, indicating faster flow setup and control plane responsiveness.
• Floodlight showed moderate performance, with better throughput than POX but worse than OpenDaylight in all tested configurations.
• POX exhibited the lowest throughput and highest RTT values, indicating slower control plane processing and higher latency.
• The performance gap between controllers was more pronounced in the mesh topology due to increased control plane complexity and inter-controller communication.