By analyzing the signal lings exchange of SIP deeply, GSPN model of protocol is established for evaluating the performance of protocol, reducing call setup time of SIP and obtaining a tradeoff between the performance and the number of buffer. Basing on the model, markov chain of model is designed. Based on properties of constructed markov chain, performance of SIP is studied and evaluated through calculating and analyzing steady-state probability and key performance indices. In the process of SIP's service the time-consuming part of transferring the state and transition is proposed. Considering the time-consuming part, the optimal size of the buffer in the place P2 of the part is computed by simulation with SPNP kit. Meanwhile, Simulation is employed to get a correlation in which average time delay of call setup is associated with the buffer of the place P2, and to evaluate the throughput of system. Performance analysis and simulation results show that the optimal size of the buffer in the place P2 is better than other size, which can reduce call setup time and improve the performance of protocol.