Fairness Improvement of Multiple-Bottleneck Flow in Data Center Networks

Quantized Congestion Notification (QCN), discussed in IEEE 802.1Qau, is one of the promising layer 2 congestion control methods for data center networks. Data center network fundamentally has symmetric structure and links are designed to have high link utilizations. So, data center flows probably pass through multiple bottleneck links. QCN reduces its transmission rate with each congestion notification feedback reception, which might cause excessive regulation of transmission rate. We have already proposed QCN with Bottleneck Selection (QCN/BS) for multicast communications in data center. QCN/BS is originally proposed for multicast communications, but it can also be applied to unicast communication with multiple bottleneck points. QCN/BS selects the worst congestion level and the transmission rate of the sending device is calculated exclusively according to feedback from the selected switch. In this paper, we preliminary evaluate QCN/BS in unicast communications with multiple bottleneck points. Our preliminary evaluation reveals that QCN/BS can resolve this excessive rate regulation problem but has new fairness problem for long-hop flow. To resolve this fairness problem, we integrates QCN/BS and our proposed Adaptive BC_LIMIT. In Adaptive BC_LIMIT, parameter BC_LIMIT is adaptively decided so that the time interval between QCN rate increase is independent of transmission rate. With rate increase interval independent of transmission rate defined in the original QCN as well as rate decrease dependent on it defined in our proposed Adaptive BC_LIMIT, convergence of fair rate allocation among flows sharing a bottleneck link is accelerated. Our simulation results show that our proposed integration of QCN/BS and Adaptive BC_LIMIT significantly improves fairness problem for unicast communications with multiple bottleneck points in data center networks. Keywords–data center; QCN; congestion control; fairness.