Fast backward predictive congestion notification for ATM networks with significant propagation delays

Abstract

A fast Backward Predictive Congestion Notification (BPCN) scheme for ATM networks is proposed, whose objective is to avoid cell loss and achieve high resource utilisation during high traffic demand. A dynamic adaptive model is proposed to capture the short term queue behaviour at each link. Based on this model, a forecast of boundaries on future queue length is generated for a time horizon projected one round trip propagation delay ahead. When potential buffer overflow is anticipated, cell loss is avoided by issuing a control message indicating appropriate constraints on the peak rate limits of controllable sources. Compared with the normal BECN scheme, the proposed BPCN has the advantage of reducing cell loss and improving network capacity.