Abstract: Doo-Sabin surfaces are the limits of recursively refined polyhedral meshes.The recursion depth n of a Doo-Sabin surface is estimated according to the error equation.Then the polyhedral mesh after n steps of subdivision is used as a substitute for the limit surface during rough machining to simplify the tool paths.The traditional G-Buffer method is extended to a strip buffer method （called S-Buffer） to generate NC tool paths for objects of large size.Furthermore,we analyze the locality of S-Buffer, adopt the weighted-response time as a performance index,and develop a preemptive and dynamic task assignment and load-balance strategy. Thus we effectively parallelize S-Buffer based on the message passing interface（MPI）.