Marzieh Khoshbakht
Mohammad Mahdi Tajiki
Behzad Akbari

In recent years, several topologies with multiple-path between each pairs of end hosts for data center (DC) networks have been proposed. However, the path diversity is shown to be not enough to improve the network performance. Researches on the DC network measurements have shown that congestion occurs even when the average utilization of links is low, which means that some of the links are over-utilized while others are underutilized and have a considerable available bandwidth. Therefore, traffic engineering (TE) is necessary for proper distribution of the network load as well as exploiting the path diversity that is provided by new topologies. Current Equal Cost Multi Path (ECMP) based approaches are not efficient in lots of cases because numerous big flows may collide on the same path. The centralized solutions depend on the ability to predict the traffic pattern, which is not effective for unpredictable traffic patterns of data centers. In this paper, SDTE, an online software defined TE approach is proposed for cloud data centers. The proposed system does not depend on the ability to predict traffic pattern or the size of flows. SDTE exploits the PEFT routing algorithm to assign weights to links. SDTE is implemented within the OpenFlow framework. The evaluation shows that SDTE performs close to the optimal routing (average deviation is about 7%).

Software Defined Networking, Data Center Routing, Multipath Routing, Traffic Engineering, Cloud Computing, OpenFlow