A Model for Endpoint Admission Control Based on Packet Loss
Endpoint admission control solutions, based on probing a transmission path, have been proposed to meet quality requirements of audio–visual applications with little support from routers. In this paper we present a mathematical analysis of a probe–based ad
- PDF / 483,134 Bytes
- 12 Pages / 430 x 660 pts Page_size
- 27 Downloads / 225 Views
Abstract. Endpoint admission control solutions, based on probing a transmission path, have been proposed to meet quality requirements of audio–visual applications with little support from routers. In this paper we present a mathematical analysis of a probe–based admission control solution, where flows are accepted or rejected based on the packet–loss statistics in the probe stream. The analysis relates both system performance to the design parameters, and the experienced probe packet loss probability to the packet loss probability of accepted flows. The goal is to provide a simple mathematical method to perform network dimensioning for admission control based on end–to–end probing.
1
Introduction
Applications that produce traffic in the Internet can be broadly classified into elastic or inelastic. Elastic flows appear mostly as a result of the transfer of digital documents, like files, web pages or multimedia downloading for local storage. These flows use TCP as transport protocol. Inelastic flows appear nowadays mainly from the streaming of audio and video over UDP without congestion control functionality. The sudden increase of video over IP on the Internet requires a better and more predictable service quality than what is possible with the available best– effort service. Audio–visual applications can handle limited packet loss and delay variation without affecting the perceived quality. Interactive communication, in addition, requires stringent end–to–end delay requirements. For example, IP telephony requires that a maximum of 150 ms one–way delay should be maintained during the whole call. The DiffServ architecture is an attempt to meet the distinct quality of service requirements of elastic and inelastic traffic, but it still lacks an important function: admission control. Without admissions control, inelastic flows can overload the network to a point where both elastic and inelastic flows suffer intolerable packet loss. Admission control has been proposed to support applications with
Corresponding author.
A. Das et al. (Eds.): NETWORKING 2008, LNCS 4982, pp. 506–517, 2008. c IFIP International Federation for Information Processing 2008
A Model for Endpoint Admission Control Based on Packet Loss
507
quality of service (QoS) requirements by limiting the network load. These proposals aim at providing QoS with very little or no support in the routers. They also share a common idea of endpoint admission control: A host sends probe packets before starting a new flow and decides about the flow admission based on statistics of the probe packet loss [1, 2], or delay and delay variation [3]. The admission decision is thus moved to the hosts and is made for the entire path from the source to the destination, rather than on a per–hop basis. Consequently, the service does not require any explicit support from the routers other than one of the various scheduling mechanisms supplied by DiffServ, and the mechanism of dropping or marking packets. In this paper we provide a mathematical analysis of the probe–based admission control (
Data Loading...
