Performance evaluation of an admission control algorithm: dynamic threshold with negotiation

An admission control algorithm for a multimedia server is responsible for determining if a new request can be accepted without violating the QoS requirements of the existing requests in the system. Most admission control algorithms treat every request uniformly and hence optimize the system performance by maximizing the number of admitted and served requests. In practice, requests might have different levels of importance to the system. Requests offering high contribution or reward to the system performance deserve priority treatment. Failure of accepting a high-priority request would incur high penalty to the system. A novel threshold-based admission control algorithm with negotiation for two priority classes of requests is proposed in our previous study. The server capacity is divided into three partitions based on the threshold values: one for each class of requests and one common pool shared by two classes of requests . Reward and penalty are adopted in the proposed system model. High-priority requests are associated with higher values of reward as well as penalty than low-priority ones. In this paper, given the characteristics of the system workload, the proposed analytical models aim to finds the best partitions, optimizing the system performance based on the objective function of the total reward minus the total penalty. The negotiation mechanism reduces the QoS requirements of several low-priority clients, by cutting out a small fraction of the assigned server capacity, to accept a new 1 This work is sponsored by the National Science Council, Taiwan, R.O.C., under the contract NSC 88-2213-E-006-057. 2 high-priority client and to achieve a higher net earning value. Stochastic Petri-Net model is used to find the optimal threshold values and two analytical methods are developed to find sub-optimal settings. The experiment results show that the sub-optimal solutions found by the proposed analytical methods are very close to optimal ones. The methods enable the algorithm to dynamically adjust the threshold values, based on the characteristics of the system workload, to achieve higher system performance.