Dynamic Topology Management in Optical Data Center Networks
Dynamic Topology Management in Optical Data Center Networks
复制标题
DOI：
10.1109/jlt.2015.2464079
复制
发表时间：
2015-10
影响因子：
4.7
通讯作者：
Yangming Zhao;Sheng Wang;Shouxi Luo;V. Anand;Hongfang Yu;Xiong Wang;Shizhong Xu;Xiaoning Zhang
中科院分区：
文献类型：
--
作者：
Yangming Zhao;Sheng Wang;Shouxi Luo;V. Anand;Hongfang Yu;Xiong Wang;Shizhong Xu;Xiaoning Zhang
近年来，云数据中心网络(DCNS)的流量呈指数增长。越来越多的用户将他们的计算需求和服务外包给DCNS。为了以可扩展的方式处理这种工作量和业务量的增加，已经提出了许多技术来向网络提供动态拓扑以适应业务量。一种典型的方案是光交换体系结构(OSA)，因为它提供了极大的灵活性。但是，OSA需要太多时间(10毫秒)来重新配置拓扑。一步重新配置拓扑不仅会给时延敏感的数据流带来很大的性能下降，而且在高吞吐量、低延迟的DCN中也会造成很大的流量损失。因此，需要一种渐进的拓扑重构方案来保证时延敏感流的性能，减少流量损失。为此，本文提出了一种拓扑管理算法(TMA)来设计这样一种渐进式拓扑重构方案。利用TMA，得到了一系列中间拓扑和布线方案。我们利用这些中间拓扑来逐步实现重新配置，同时保持拓扑连通性，减少拓扑重新配置期间的总流量损失。通过大量的实时跟踪驱动的仿真实验，我们发现TMA能够以很小的开销减少拓扑重构过程中的流量损失。
In recent years, there has been an exponential growth in the traffic of cloud data centers networks (DCNs). More and more users are outsourcing their computing demands and services to DCNs. To handle such increases in the workload and traffic in a scalable manner, many technologies have been proposed to provide dynamic topology to the network so as to adapt to the traffic. One representative scheme is the optical switching architecture (OSA) as it provides tremendous flexibility. However, OSA requires too much time (10 ms) to reconfigure the topology. Reconfiguring the topology in one step not only brings much performance degradation to the delay sensitive flows, but also incurs large traffic losses in high throughput low latency DCNs. Therefore, a progressive topology reconfiguration scheme is required to guarantee the performance of delay sensitive flows and reduce the traffic loss. To this end, we propose a topology management algorithm (TMA) in this paper to design such a progressive topology reconfiguration scheme. With TMA, a sequence of intermediate topologies and routing schemes are derived. We leverage these intermediate topologies to realize the reconfiguration progressively, while maintaining the topology connectivity and reducing the total traffic loss during the topology reconfiguration. Through extensive real-trace driven simulation experiments, we find that TMA can reduce the traffic loss during topology reconfiguration with little overhead.