AbstractNetwork services with deterministic guarantees are based on a worst—case description of user—generated traffic. When designing a policing and scheduling algorithm for guaranteed services on the Internet, accuracy of description of the traffic profile has to be traded with simplicity of implementation. The result of this trade off is often expressed as the number of token buckets required by the service along with the choice of their parameters. The GS type of service proposed by the IETF uses two token buckets both for characterizing the traffic and for policing it. The choice of using only two token buckets is primarily driven by policing costs. In this paper we propose a method that allows the number of token buckets used for characterizing the traffic to be greater than what is actually needed to police it. This means we can obtain an accurate profile of the traffic while keeping policing simple. The method consists of computing a profile of the traffic which involves a number of token buckets of the order of ten, and then doing the policing using only the first token bucket, plus another one which is chosen depending on the delay requirements of the receivers. This paper shows that with this simple enhancement we obtain a guaranteed service whose performance closely approaches the theoretical limits of services with deterministic guarant