近几年，“云”已经从IT业界的新兴技术转化为广为接受和部署的解决方案。随着云用户数量的激增，数据中心的规模和流量正在以爆炸性的方式增长。据思科预测，2016年数据中心流量会达到6.6 zetabytes (10^27 字节) ，为2011年的4倍。现有的任何网络解决方案都无法处理如此海量的数据，设计下一代大规模、超高速数据中心网络的需求变得极为迫切。本申请课题把数据中心网络作为一个复杂的、整体的系统科学进行研究，解决了一系列相互关联、相辅相承的科学问题。首先深入了解和探索在提供完全平分带宽条件下拓扑结构的硬件复杂度最优化策略，分析怎样在海量交换节点间通过高速路由获得负载平衡，探讨节点或链路发生故障后能持续提供QOS保证的快速恢复技术，并研究了大传播时延环境下具有高吞吐量的输入输出控制方法。这些研究成果为设计下一代大规模、超高速数据中心网络提供了必要的理论支撑和合理的技术方案。

In recent years, the "cloud" has been transformed from the new emerging IT technology to a widely deployed and accepted solution. With the surge in number of cloud users, the traffic volume in data centers is growing at an explosive way. Cisco predicts that the traffic in data center will reach 6.6 zetabytes (10^27 bytes) in 2016, which is about four times than that of 2011. Any existing network solution cannot handle such vast amounts of data. The need for designing the next generation of large-scale and ultra-high-speed data center network becomes extremely urgent. In this proposal, we discuss the data center network as a system and solve a series of interrelated scientific problems. Specifically, we guarantee the full bisection bandwidth with the minimal hardware complexity. We dedicate a fast routing algorithm to provide the perfect load balancing among a great number of switches. We propose a rapid recover technology that can provide QoS when encountering node or link failures. Last but not least, we design an efficient scheduling algorithm under large propagation delays. With no doubt, these contributions will be the references or guidelines for the next generation of large-scale and ultra-high-speed data center networks.