Implementation and Experimental Evaluation of Flow Diffusion Algorithms for Folded Clos Networks

Abstract

Folded Clos networks (FCNs) play a significant role in constructing high-performance data center networks. To fully utilize the high bandwidth of FCNs, traffic load must be uniformly diffused between links. To achieve this, a method called the rebalancing algorithm was proposed. Previous studies have demonstrated the effectiveness of this algorithm through theoretical analyses and computer simulations. However, its practical feasibility has not yet been elucidated. Moreover, its performance has not been confirmed in a physical experimental environment. This study demonstrates the implementation of the rebalancing algorithm on switches built on PCs using software tools supported by the Linux OS. This confirms that it is relatively easy to implement the rebalancing algorithm. In addition, the performance of the implemented rebalancing algorithm was evaluated and compared with that of its simplified version, called the balancing algorithm, as well as a conventional method, through experiments. According to the results, the rebalancing algorithm is more advantageous in avoiding traffic congestion for heavy traffic than the conventional method. It was also confirmed that the balancing algorithm performs as effectively as the rebalancing algorithm, with less processing load.