An important feature that differentiates moving object database (MOD) from other databases is that MOD can support queries for locations recorded not only at the sample time, but also at time between samples and even at time in the future. The most important issue in MOD research is to build models of moving objects?mobility and their locations update. The advance in mobile Internet and location-based services requires appropriate models be built for popular dumb terminals, such as cell phones, which can only be located by auxiliary facilities like GSM/CDMA networks. Considering that users of such kind dumb devices often move on the road network, a novel model for these moving objects to calculate their velocities in the past and future and to update their locations with varied sample time based on road network is presented in this paper. Compared with traditional velocity models, the proposed model can reduce the location prediction errors effectively. And when comparing with the model of updating locations with uniform sample time, the proposed model can reduce the communication traffic between moving objects and locating facilities with the same average prediction error.