MBAC Robustness to Traffic Variations: A Simulation Study

Measurement based admission control (MBAC) has long been recognized as the best method for providing stochastic service guarantees. Since MBAC relies on measurements, the actual performance of MBAC can only be established through simulations or on real networks. There are two main performance measures: 1) how well the MBAC can actually meet the QoS target specified by a flow; and 2) how well the MBAC can utilize the network resources. The main work in the literature has mostly been on finding optimal MBAC algorithms that increase network utilization. In this simulation study we focus on an MBAC's ability to meet the QoS target under variations in the number of flow attempts within an interval and variations inflow lifetimes. We introduce an ideal admission controller based on analytical values and show the importance of using this ideal when assessing the performance of an MBAC. Most MBAC algorithms rely on tuning parameters to help in the decision making. Tuning these parameters is a difficult task, and a setting that gives excellent performance under one scenario, may give a very pessimistic or too optimistic performance in another scenario. We show that by adding two policies, the so called pessimistic policy together with a back-off policy, the reliance on tuning parameters diminishes. With these two policies in place we show that an MBAC can perform close to the ideal admission controller under various traffic loads without relying on explicit tuning parameters