PDFR: Path Delay Based Flow Rerouting in Software-Defined Networks Hnin Thiri Zaw*, Aung Htein Maw** University of Computer Studies, Yangon*, University of Information Technology** h.thirizawucsy@ucsy.edu.mm *, ahmaw@uit.edu.mm ** Abstract This paper presents path delay based flow rerouting (PDFR) algorithm for traffic engineering in software-defined network (SDN). The goal of PDFR is to reroute the elephant flow from ongoing path to least delay path in order to improve network performance. The evaluation results prove that PDFR has not only 21%~71.32% throughput improvement and 14.44%~49.25% flow completion time (FCT) reduction for elephant flows, but also 6%~40% FCT reduction in mice flows (short flows) compared with the single path method. Since PDFR works based on the delay metric of path, the future work will consider not only delay but also traffic loads of the path in order to achieve better performance results. Keywords: PDFR, Software-defined network (SDN), elephant flow, mice flow, rerouting. 1. Introduction As a well known data center network topology, the fat-tree topology contains various paths between hosts, so it can give higher accessible data transmission than a single path tree with a similar number of nodes. It is normally a 3-layer various leveled tree that comprises of switches on the edge, aggregate, and core layers. The hosts in the lowest layers are connected directly to one of the edge layer switches. The aggregate layer switches interconnects multiple edge layer switches together. All of the aggregate layer switches are connected to each other by core layer switches. Core layer switches are also responsible for connecting the data center to the Internet. For fat-tree topology, maximizing network throughput and minimizing transferring latency are two critical targets. So as to achieve them, there exist both equipment and programming approaches. New technologies have changed because the nature of networking has been more and more complicated. For example, the cloud computing and massive data centers demands have made effective networking much more complex. To adapt to these requests, network administrators need their systems to be smarter and they also need to have the capacity to better control and manage them. Accordingly, software-defined networks (SDN) become the new emerging infrastructure to address these issues. The software-defined network (SDN) architecture, which decouples the data plane and control plane, allows network administrators to program the behavior of their networks with an external SDN controller, which has the ability to change the forwarding behavior of the network element directly [1]. The SDN is a network infrastructure with high adaptability and network operators can manage greatly the SDN-enabled switches by the programmability. Due to the combination of virtualization and solidification, network operation costs can be eliminated, by