Network tomography provides the ability to employ the end-to-end measurements to infer the network-internal link performance parameters indirectly without requiring cooperation from the intermediate elements of the network. As a significant alternative to network measurements to be able to guide the network management as well as the network optimization, network tomography receives a plenty of attention both in academia and industry. This survey is based on an extensive collection and reference of research works at home and abroad. First, the measurement schemes exploited by the network tomography are summarized. Next, the corresponding tomographic approaches are divided into two classes regarding at what granularity they describe the link's performance: the quantitative parameter estimation and the qualitative parameter estimation. Then according to inference problems of the different parameters, a general analysis of the existing algorithms is conducted. Lastly, future research areas and potential applications are suggested.