Modeling and performance evaluation of OpenFlow switches using a MAP/PH/1/n queueing model

Abstract

Software-Defined Networking (SDN) is a paradigm shift in network architecture. It decouples the control plane from the data plane to enable centralized network management and programmability. While Software Defined Networks (SDNs) offer significant advantages by efficient traffic management, it also introduces complexities that require comprehensive network modeling to predict and optimize network behavior before actual deployment. Queueing models provide a mathematical framework for analyzing and predicting how data packets behave as they traverse network devices. This paper presents a discrete-time MAP/PH/1/n queueing model to assess the performance of SDNs in handling complex and bursty traffic patterns. The model integrates packet processing at different switch components, including the switch buffer, ingress processing unit, and egress processing unit. It utilizes a finite buffer queue model with Markovian Arrival Process (MAP) and Phase-Type (PH) service times to capture data transmission behavior at an OpenFlow switch. The matrix geometric method is employed to calculate steady-state probabilities, which helps in evaluating Quality of Service (QoS) metrics such as average delay, throughput, and blocking probabilities. In addition, the mathematical model formulates performance measures, including probabilities for packet forwarding, packet drop, and packets redirected to the controller. We validated our model’s outcomes by conducting packet-based simulations using Mininet and the Ryu controller. The graphs obtained from both the mathematical model and the packet simulations demonstrate qualitatively similar behavior of the OpenFlow switch across different traffic rates, buffer sizes, and service rates.