分布式文件系统在管理海量数据的同时，也带来大量并发的网络传输，容易引发或加剧局部网络的拥塞，进而导致数据中心内部网络的不均衡使用。研究发现分布式文件系统读写操作中所采用的数据块选择策略，存在较大的优化空间。关键问题在于如何让系统实时感知其所处网络的状况及变化，在挑选目标数据节点时避开拥塞的网络区域，从而确保网络资源的负载均衡。为证明此设想，在不改变HDFS高容错性和高吞吐量等已有特性的情况下，研发网络度量框架DFS-NetEye，探索读写调度算法通用测试框架，尝试建立网络负载均衡评价体系。

When distributed file system managing big data, it also brings massive parallel data transmission that may lead to regional network congestion and imbalanced usage. .Research has found that, the chunk selection strategy employed by distributed file system can be optimized. The key point is to measure the network status of a distributed system in a real-time manner, and to choose reasonable data node in response to each incoming read/write request. .To prove this idea, this project proposes DFS-NetEye, a network measurement and scheduling system. We explore the possibility of building an open scheduling algorithm test framework, as well as establishing network workload evaluation architecture.