An integrated admission control based on measurements in ATM networks

Abstract

We developed a novel scheme to maintain QoS (Quality of Service) requirement for different traffic classes in the dynamic environment through full utilization of statistical multiplexing. Previous designs on measurement-based admission control mainly focused on strategies that consider the worst case traffic source model to guarantee QoS bounds for all connections. This can result in overallocation of resources. The proposed mechanism not only allows statistical multiplexing gain to be exploited, it also simplifies the estimation process. An accurate formula for the cell loss probability which is combined with measurements, is presented for a new traffic model. Furthermore we enhance the performance of this mechanism through real-time measurements of traffic and monitoring of QoS.

Introduction

Asynchronous transfer mode (ATM) has been the hottest topic in the networking community and proposed by the International Telecommunications Union (ITU) as the transport mechanism of choice for B-ISDN (Broadband-Integrated Services Digital Network). However, there are a few issues which must be solved to implement efficient and robust ATM networks. To maximize the merits of ATM, designers have been faced with a wide range of problems, varying from the implementation of mechanisms for efficient congestion control, call admission and routing to the development of lightweight transport protocols. In particular, as pointed out in many researches, traffic control is the first aspect to be considered [15]. Traffic control remains, however a controversial problem in practice: how to provide network resources (Call Admission Control: CAC) and how to monitor the conformity (Usage Parameter Control: UPC). One of the reasons that traffic control is difficult in ATM networks is the diversity of traffic characteristics and QoS requirements. Adaptability to unknown and variable traffic demand and flexibility for introducing new services are more important in ATM networks than in conventional networks.

There have been numerous studies on call admission control. In most of these studies, the approaches depend on restrictive and precise assumptions on traffic models, or involve complex and computationally expensive procedures of matrix computations. These approaches, while representing the state-of-art in modeling and analysis, are hardly practical for real-time admission control schemes. Therefore, many analyses have adopted traffic measurements for admission control [1], [2], [3], [4]. However, relying on measured quantities raises estimation error problem that impairs developing robust control.

In this paper, we propose an alternate approach to efficient resource allocation and providing strong QoS guarantees. This approach integrates measurement process with cell loss probability estimation to prevent the overallocation of bandwidth and reduce the estimation error from the control based on network measurements alone. Since cell delay can often be controlled within a desired bound by adjusting the buffer size, we mainly concentrate on cell loss as the QoS measure of interest [5], [6], [7], [8]. The instantaneous cell loss rate of a new call can be calculated from simple formula which implements an implicit measurement procedure. And each traffic source can be modeled as on/off process as in many approaches [2], [9], [10]. In order to achieve higher bandwidth utilization, we use alternative two-state model which reflects statistical multiplexing gain efficiently.

We emphasize that the main advantages of the CAC developed here are its high link efficiency and low computational overhead that makes it practical for real time admission control. The performance of the developed algorithm is illustrated via analysis and simulation.