Traffic classification and identification is the key part in network security industry. It can identify the types of the Internet traffic, and detect abnormal ones based on the features of network flows. Nowadays, due tothenon-transparency and complexity of the packets, machine learning methods are widely adopted to identify the abnormal traffic. As a classic supervised learning algorithm, SVM performed well in traffic identification, including the speed of training and predicting as well as the accuracy. However, with the amount of network traffic being larger, stand-alone SVM cannot meet the requirements and be difficult to deal with the large scale of network traffic. So, in this paper, we used parallel computing on spark to accelerate and fast deal with model training and predicting. At last, the comparison of training time, prediction time and accuracy with stand-alone SVM and SVM on spark will be given. Besides, the analysis in detail will be also presented.