Saito, H., "Adaptive CACs Using Bayesian Performance Estimation," ICCCN'95
This paper investigates adaptive CACs using Bayesian performance estimation. The performance is estimated through Bayesian regression analysis, in which the monitored performance and an a priori estimate are given by conventional techniques such as queueing theory and simulation. Three adaptive CACs are studied: a CAC based on a log-transform of the cell loss ratio (CLR), a CAC based on a maximum queue length during a fixed period, and a CAC based on CLR. Numerical examples show the following results: oscillation of VCs in a VP may occur and may be worse when the measurement period is longer; Initial variance is important for accurate estimation; CACs using CLR and the log-transform of CLR can not easily assure the CLR objective; the CAC using the maximum queue size performs well and has rapid convergence; the monitored performance data improves estimates even when the performance is monitored with different conditions than the target condition.