When microscopic traffic simulation is used to predict the traffic jam and to evaluate the transportation planning program, the speed of simulation is important. In the distributed microscopic traffic simulation system (DTMS), the configuration of terminals is not only the key to ensuring the speed-up of the simulation, but also the basis of reducing the cost of the simulation. In this paper, we build an efficient time overhead model of DTMS and a greedy-growing algorithm (OCTA) used to optimize the configuration of terminals. The configuration of terminals mainly includes the computing of the number of needed terminals, and the distribution of sub-network partitioned from the road network. Based on the top-down time overhead model of DTMS, the OCTA algorithm partition the road network and distribute each sub-network to terminals to finish the configuration of terminals. Experimental results show that the model and the algorithm can compute the number of terminals of simulation under the condition of the needed speed-up of simulation. This algorithm has been applied in our simulation platform.