快速准确侦测和调度大象流可以在提高网络性能的同时优化网络资源。但是由于传统的网络架构缺乏通用机制对大象流的定义、侦测、和调度进行全局管理，通过侦测调度大象流优化网络整体性能和资源的目标无法最大化。SDN(软件定义网络)架构具有逻辑集中的控制平面。其通用强大的数据流管控机制可以对大象流的定义、侦测、和调度的进行全局管理, 但现有的侦测和调度方法以控制平面生成的数据流统计信息为基础，网络消耗大。本项目另辟蹊径，提出了一种新型的大象流侦测方法。该方法无须收集转发平面的统计信息。相反，它让控制器直接统计转发平面发送到控制平面的数据分组，通过灵活细粒度地调整大象流流表项的有效类型和时间，在迅速准确侦测调度大象流的同时，优化了网络的整体性能和资源利用。该方法无需部署专用终端主机，升级现有网络设备，具有速度快精度高，适应动态变化的网络和应用环境的特点，可以直接用于5G网络中。

Accurately detecting and scheduling elephant flows in their early stage has a benefit to significantly improve network performance and optimize network resources. However, the conventional network architecture lacks general mechanisms to define, detect, and schedule elephant flows in a global manner, leading to a difficulty in maximizing the benefit. Software Defined Networking (SDN) architecture has a logically centralized control plane. With unified and flexible mechanisms in controlling the forwarding of flows, SDN can define, detect, and schedule elephant flows globally. However, previously proposed approaches in detecting and scheduling elephant flows are based on the network statistics maintained by the switches and routers in the data plane, resulting in a huge network bandwidth consumption, and significantly decreasing the network performance. Therefore, we propose a novel approach in detecting and scheduling elephant flows. This approach does not need to collect the network statistics generated by the switches and routers in the data plane. In contrast, it lets the controllers in the control plane count the statistics of packets of flows forwarded to the control plane. By flexibly and finely adjusting the timeout of flow entries of flows, the proposed approach can accurately detect and schedule elephant flows in their early stage without degrading the network performance and resource usage. The proposed approach does not need to deploy dedicated hosts and update the existing switches and routers. It can adapt to the changes of dynamical networking environments, and readily apply to the data centers of the upcoming 5G systems.