Resource requirements for ABR explicit rate flow control: Deterministic and probabilistic analyses

Citation
I. Hsu et al., Resource requirements for ABR explicit rate flow control: Deterministic and probabilistic analyses, TELECOM SYS, 16(3-4), 2001, pp. 297-313
Citations number
10
Categorie Soggetti
Information Tecnology & Communication Systems
Journal title
TELECOMMUNICATION SYSTEMS
ISSN journal
10184864 → ACNP
Volume
16
Issue
3-4
Year of publication
2001
Pages
297 - 313
Database
ISI
SICI code
1018-4864(2001)16:3-4<297:RRFAER>2.0.ZU;2-S
Abstract
ATM Available Bit Rate (ABR) service is intended to offer low cell loss for non-real-time data sources that can respond to closed-loop flow control. A TM Forum Traffic Management Specification Version 4.0 defines the various p arameters used in the ABR flow control, as well as the source, destination, and switch behaviors. However, the switch designers and service providers are free to choose the method of congestion control to implement and the AB R Quality of Service (QoS) objective to offer. This paper addresses the int eraction among the flow control algorithm, the switch resource requirements , and the resulting QoS characteristics. In this paper we propose and evaluate an Explicit Rate (ER) algorithm. The objective of this algorithm is to maintain the total buffer occupancy of al l ABR connections to be close to a given threshold. By maintaining a non-ze ro queue, the ABR service can achieve a high utilization. The switch period ically determines its desirable ER value, based on the available capacity, the ABR buffer occupancy, and the number of active ABR sources. We develop analyses that relate ABR resources to QoS objectives for this al gorithm. The first approach is a deterministic, conservative analysis. It p rovides formulas for determining the ABR buffer and capacity requirements t hat can achieve zero buffer overflow. The second analysis determines an upper bound on the buffer overflow probab ility when the above requirements are not met. The result is most effective when the number of active sources is a small fraction of the total ABR con nections. Numerical examples show that by slightly relaxing the loss constr aint, resource requirements can be significantly reduced.