Integrated bandwidth and congestion management in ATM networks using a simple learning algorithm and a bucket bank

A simple learning-algorithm-based method for dynamically sharing a pool of bandwidth (BW) by multiple categories of traffic in asynchronous transfer mode (ATM) networks is proposed. For constant bit rate (CBR)-type traffic, only the maximum of the consumable BW is defined, whereas for the other categories, a minimum portion of the BW is guaranteed when they are sufficiently active. The remaining portion of the BW is dynamically shared on the basis of pre-assigned priority level, discard eligibility, and current activity of the cells of a specific category. Functions for (1) fast reward fast penalty (FRFP), (2) fast reward slow penalty (FRSP), (3) slow reward slow penalty (SRSP), and (4) slow reward fast penalty (SRFP) are proposed. Their impact on the achievability of dynamic sharing of BW is investigated