A family of measurement-based admission control algorithms

In this paper we introduce our concept of the requirements of an e cient admission control algorithm and propose new algorithms. Our methods need only aggregate tra c measurements and simple FIFO scheduling. We take minimal assumptions on the pattern of the tra c. Our framework de nes a family of measurement based admission control algorithms of which we discuss three key members. We give e ective bandwidth formulae for bu ered and bu erless systems with token bucket and peak rate limited sources. We also show how the utilization can be improved by measuring the variance of the tra c rate while avoiding the limitations of the Central-limit theorem. The theoretical background of our work is the e ective bandwidth de nition introduced by Gibbens and Kelly which has its roots in the Cherno bounds.