Abstract. The support of real-time traffic in class-based IP networks requires reservation of resources, accompanied by admission control in order to guarantee that newly admitted real-time traffic flows do not cause any violation to the Quality of Service (QoS) perceived by the already established ones. In this paper we highlight certain issues with respect to bandwidth allocation and admission control for supporting real-time traffic in class-based IP networks. We investigate the implications of topological placement of both bandwidth allocation and admission control schemes. We show that their performance depends highly on the location of the employed procedures with respect to the end-users and the various network boundaries. We conclude that the strategies for applying these schemes should be location-aware, because their performance at different points in a class-based IP network can be different and can deviate from the expected performance. Through simulations we also provide a quantitative view of these deviations. 1

Multiclass IP networks open new dimensions and challenges on active monitoring as efficient strategies of in-band probing are required to sense each class performance without causing noticeable side-effects on real traffic. In our study, we provide new insights on how to perform efficiently active monitoring in these networks, suggesting the use of light and multipurpose probing streams able to capture simultaneously the behavior of multiple QoS metrics of each class. Considering oneway -delay, jitter and loss metrics, we explore different spatial-temporal characteristics of probing, focusing on finding patterns adjusted to each class measurement requirements. We demonstrate that commonly used probing streams fail to capture these metrics simultaneously and we propose novel colored probing patterns able to increase multipurpose active monitoring efficiency. As test environment, we consider a diffserv domain where admission control resorts to feedback from edge-to-edge active monitoring to dynamically control hard real-time, soft real-time and elastic traffic classes. Comparing graphically and statistically the probing and passive measurement outcome of each class, the obtained results show that despite being difficult to match the scale and shape of multiple metrics, a single and properly colored probing stream can capture close and simultaneously the behavior of one-way-delay, jitter and loss, for low in-band probing rates.

This paper explores the use of edge-to-edge active monitoring to control simultaneously multiple QoS parameters in multi-service IP networks, while reducing the effects of intrusion on real traffic. Considering a multi-class domain where traffic is controlled at network boundaries based on feedback from on-line measurements, the present work is centered on obtaining adequate per class in-band probing streams so that each class behaviour is correctly captured, even if more than a QoS metric is under control. In this way, we investigate distinct properties of probing patterns and cross-check probing and passive measurement results in order to assess and tune probing effectiveness. To enhance probing ability to sense multiple metrics, we explore Active Queue Management effects on probes and their different probability of reaching the network boundary.

Abstract not found