Achievements of network traffic simulation provide realistic, feasible and accurate methods of measuring bandwidth, latency, jitter and so on. However as the development of networking technology, traditional manners are becoming insufficient to provide high bandwidth matching the real network, nor accurate statistics characteristic such as self-similarity. In this paper we propose the Network Traffic Simulation platform based on Network Processor (NTS-NP). The platform has achieved: 1) high bandwidth and real-time traffic generation; 2) high precision of self-similarity of the flow, which is now considered to be one of the basic characteristics of network traffic. In our platform, Cavium OCTEON CN3860 is chosen as the core in order to gain the ability of high speed packets assembling and transmitting, and direct control of the interface hardware. ON/OFF aggregation model is implemented to generate self-similarity and a discrete-event driven based algorithm is proposed to improve the traditional implementation of the model. We estimate the Hurst parameter of synthetic traffic by R/S statistics method and wavelet estimator. The validation of the platform is also empirically discussed and proved.