近年来，很多在线服务供应商建立数据中心来提供在线搜索、社交网络等大规模网络应用服务。由于这些网络服务产生了具有不同流量特性和应用需求的异构混合流，现有的数据中心网络面临着一个重要挑战：如何在数据中心网络异构流传输场景下提升网络应用的服务性能。本项目分析数据中心网络异构流的特性，探索一套负载均衡机制，来满足各应用的传输需求。主要研究内容包括：（1）融合全局和本地的拥塞程度信息，设计负载均衡粒度调节方法；（2）基于异构流的传输模型，研究负载均衡控制机制，满足不同应用的需求；（3）面向负载均衡多路径传输场景，设计编码算法和缓存管理策略，以降低流完成时间；（4）在交换机的显式通告辅助下，设计乱序区分和路径感知的拥塞控制算法以提升传输效率。通过解决这些科学问题，保证数据中心的高效、公平和可靠的网络传输，对推进数据中心的研究发展有重要意义。

In recent years, many online service providers construct their data centers to handle the large-scale network applications such as web search and social networking. Since these applications generate a diverse mix of flows with different traffic characteristics and performance requirements, the data center networks are currently facing a severe challenge in improving the service performances under the heterogeneous network traffic. This project aims to explore the traffic characteristics of heterogeneous network traffic in data center network and propose the load balancing schemes to meet the requirements of various applications. Our works address several fundamental problems including: (1) Synthesize the local and global congestion information and adjust the granularity of load balancing; (2) Study the transport model of heterogeneous flows and propose the load balancing scheme to meet the requirements of different applications; (3) Design the data coding algorithm and buffer management scheme in the multiple paths transmission under load balancing to reduce the flow completion time; (4) Explore the method to differentiate the packet reordering from the packet loss, and design the path-aware congestion control to improve the efficiency with the explicit notifications from the switch. Solving these problems can provide the effective, fair and reliable transport in data center network, thus substantially advance the research of data center.