A novel approach for smooth bandwidth allocation (SBA) is proposed in this paper that uses multiple state decision mechanism to eliminate large fluctuations in the allocated bandwidth. This algorithm is a dynamic bandwidth allocation which predicts the requirement bandwidth for the input traffic according to its arrival traffic and the SLA constrains. Simulation experiences show that SBA algorithm decrease 20%-45% allocated bandwidth variance in contrast to similar methods, also decrease number of long fluctuation to 30-45% similar methods while algorithm precision is high and queuing delay is low. Different self similar traffic models are used in this experience.