Performance analysis of admission control algorithms based on reward optimization for real-time multimedia servers

We propose and analyze a class of admission control algorithms based on reward optimization for multimedia servers designed to provide on-demand services to clients in an environment where workload characteristics of clients can change dynamically. These admission control algorithms are developed based on two strategic choices, namely, "deterministic" or "best-effort" regarding quality of service (QoS) control, and "priority-reservation" or "no priority-reservation" regarding reservation control. We first formulate the design of admission control algorithms for real-time multimedia servers as a reward optimization problem, with the "reward" referring to the value which the system receives after servicing prioritized clients based on the QoS requested and delivered. Then, we derive closed-form solutions for the reward rate expressions which the system can possibly obtain as a result of applying these admission control algorithms and validate the analytical results with a simulated VBR video server. A physical interpretation is given for the best combination of strategic choices under which the system can obtain the best reward as a function of workload and cost characteristics of the clients. We demonstrate that the reward value which the system receives under our proposed admission control algorithms at optimizing conditions can be much higher than those under traditional admission control algorithms which do not consider reward optimization. (C) 1998 Elsevier Science B.V. All rights reserved.