Search

A majority of traffic engineering techniques aims to minimize network congestion in SDN, but they commonly have considered some aspects of the flows and their demands for evaluating applications in SDN. This paper focuses on a scalable traffic engineering scheme for efficiently mapping traffic demands to paths in SDN datacenters. In this regard, an optimal solution is introduced to schedule elephant flows across paths to mitigate network congestion and increase load balancing in SDN. Nevertheless, optimal flows scheduling mechanism is still not applicable because the required time to calculate the optimal paths is not tolerable in SDN. We present a new decomposition technique for limiting the search space based on the linear programming method to solve the problem in a reasonable time. Since the mice flows have the sensitive time requirements, some paths are reserved for immediately forwarding these flows. The proposed method achieves reasonable performance regarding computation time compared with state-of-the-art solutions devoted to this topic. The results also show that the proposed traffic engineering method is effective to control network congestion and provide load balancing in SDN datacenters. [ABSTRACT FROM AUTHOR]

Abstract: Data center has recently emerged as a common infrastructure that includes numerous servers and host diversity of applications and cloud‐based services. In this paper, a dynamic method is proposed for traffic engineering in software‐defined data center networks helping to design an optimal demand‐path mapping with a tolerable computational complexity. To achieve the aim of this study, we used a linear programming to define the problem statement and then modeled the problem as a Path‐based multicommodity flow. To reduce time complexity, first, a decomposition technique is used to divide the problem model into smaller subproblems, which are then solved simultaneously by applying parallelizing techniques (ie, multiple‐core computing and OpenMP). Next, a suitable iterative algorithm is utilized to quickly converge subproblem solutions leading to an answer to the master problem. Simulation results show that solving time is considerably improved compared with a number of reported schemes devoted to this topic. In addition, implementing the proposed method serially yields an optimum solution, which takes significantly shorter time to reach than that of the generic (the LP model that is not decomposed) approach. [ABSTRACT FROM AUTHOR]