Delay-Aware Elephant Flow Rerouting in Software-Defined Networking (SDN)

Abstract

Conventional tree topology networks and single shortest path routing will create congestion along overloaded links. Therefore, load-balanced path selection is necessary to reduce congestion and enhance the performance of the application. In order to solve network congestion problem and to balance the network load, Equal Cost Multipath (ECMP) algorithm is used to handle traffic flows. However, ECMP algorithm lacks the ability to schedule the traffic flow dynamically. Moreover, ECMP does not consider any current network and traffic conditions. Software-Defined Networking (SDN) is an advanced network infrastructure offering cost-effectiveness and high interoperability for end-user control and network programmability in networks. The primary difference between a conventional networking and SDN networking is decoupling data plane and control plane in SDN based networks. In SDN infrastructure, the whole network is centrally managed by a dedicated controller, which interacts with network switches using OpenFlow protocol. The network congestion detection and dynamic scheduling mechanisms can be implemented by using OpenFlow protocol in SDN infrastructure. Traffic Engineering (TE) in SDN is a critical application that is attracted tremendous amount of research to find optimized routing decision based on analysis of network and traffic conditions. Following traffic engineering in SDN, most of literature reviews differentiate the traffic flows in networks: small-volume (shortlived) flow which is referred to as mice flows and large-volume (long-lived) flow which is referred to as elephant flow. Among these two types of flows, elephant flows can affect the network performance due to consuming large amount of resources. Therefore, differentiation types of flows and scheduling elephant flows are important issues in TE. A large number of elephant flow scheduling approaches in traffic engineering have been proposed. At first, various examples of existing traffic measuring and rerouting methods are analyzed, and then their advantages and disadvantages are compared, based on a number of significant criteria such as the improvement of network throughput, the reduction of network delay and packet loss. Recent research on traffic engineering has focused on collecting flow statistics, byte counts and bandwidth utilization; however, a majority of the solutions does not address delayrelated issues.iii This dissertation presents a study towards an elephant flow scheduling approach for various types of traffic such as TCP, UDP, HTTP and FTP. To this end, a new traffic management method is proposed, that is, Delay-Aware Elephant Flow Rerouting (DAEFR), aiming to minimize the static ECMP's flow collision problem and network congestion, thereby optimizing network throughput and, reducing flow completion time and packet loss. Since the proposed method is flow-based rerouting method, the risk of packet reordering can also be reduced without additional network overhead. The proposed design for dynamic multipath-based traffic management approach in the SDN comprises five main tasks: detecting elephant flow, computing shortest paths, estimating end-to-end delay and bandwidth utilization, calculating least cost path, and rerouting traffic flow from the ongoing path to the uncongested path.