Dynamic resource allocation based on measured QoS
Sanjeev Rampal, Douglas S. Reeves, Yannis Viniotis · NCSU Libraries Repository (North Carolina State University Libraries) · 1996
Most methods for guaranteeing quality of service (QoS) in packet-switched networks require a static characterization of the user's trac, and allocate resources accordingly.This process is inconvenient for unsophisticated users, and can result in overallocation of resources.We address the problem of automatically determining the minimal resources necessary to satisfy a specied Quality of Service (QoS) measure using dynamic techniques.Specically, w e study a dynamic control algorithm which determines the minimum bandwidth needed to satisfy a specied average cell loss probability requirement f o r a g i v en source.This algorithm (referred to as REQS for Resource-Ecient Quality o f Service) measures the actual cell losses, and uses this information to dynamically vary the bandwidth allocation.Experimental results show the algorithm converges quickly to the desired solution.The eect of the measurement frequency, source burstiness, buer size and loss specication on the convergence time of the algorithm are all investigated.The algorithm is found to be robust, converging quickly and accurately under most conditions.We demonstrate the bandwidth savings attainable by this approach over other techniques such as the equivalent capacity f o r m ulas.The applicability of the method for other QoS measures (such as queuing delay percentiles) is also demonstrated.Some of the applications of this technique include control of rate-enforcing servers, trac shapers and call admission control in ATM networks.