Multilevel QOS And Traffic Models For

Multilevel QOS and Traffic Models for

Dynamic Management of Multimedia Connections

Ali Marefat

Laboratoira PRISM

University of Versailles, France

marefat@prism.uvsq.fr

http ://www.prism.uvsq.fr/~marefat

Abstract

Over the past few years extensive research works on the area of Quality of Service provisioning for multimedia applications have been done and Quality of Service based architectures are being developed. More investigations are necessary to study the impact of interactions between multimedia applications in the context of a global network management policy. Between the best effort environments offering no guarantee of performance from one side, and new ATM based static connections with total guarantee of service but to few flexibility for providing dynamic management of network resources, a tradeoff has to be found. Multilevel modeling is proposed in this paper as offering the necessary support for specification of application’s QOS and a guideline for establishing and managing multimedia connections. The resulting traffic contract is presented as a management scenario to deal with network fluctuations. To achieve higher degree of efficiency in resource utilization, virtual allocation and released-on-demand allocation concepts are proposed.

Keywords

QOS specification, Connection management, Resource reservation.

1. Introduction

With the development of ATM networks and the Internet’s new generation of protocols, new and high quality multimedia applications are appearing on the networks. More and more resource demanding, these sensitive flows has to be handled with more efficient mechanisms. To do so Quality of Service (QOS) based network architectures are being developed (see [1] for an overview on these architectures). While the objective of these researches is to provide efficient QOS to applications, new problems arise when these applications have to access to resources and to be granted QOS, without blocking other demanding applications. Given the type of the application and users, more or less flexibility could be associated to these applications. The idea is to see how this flexibility can be expressed and exploited to offer more efficient resource sharing and provide mechanisms to deal with application’s QOS requirements modifications during sessions and network’s traffic load variations.

The guide line in this paper is to try to propose a model where from one side enable an application to specify it’s desired QOS, and from the other side provide a prediction scenario of the application’s behavior, to be used to support dynamic connection and QOS management mechanisms.

In this view advantages in using multilevel modeling for QOS specifications and traffic description are studied. In section 2 we will see what facets in QOS are more important and what kind of guarantees could be offered to applications. In section 3 a multilevel QOS model for specification of the QOS is proposed and we will see how this model would be mapped on a multilevel traffic contract used for, dynamic management of connections and resources allocation.

2. Multimedia networking requirements

Multimedia networking and more particularly offering QOS to interactive applications is a challenge for new network technologies. We will first see what QOS parameters are most important for users and what service guarantees could be offered to provide the requested level of performance.

2.1. User requirements

The presentational quality at the interface between the application and user, where multimedia data are displayed, has two aspects : quantitative aspect which characterizes the spatial and temporal quality of the presentation : frame rate, resolution and color quality for a video flow for example, and qualitative aspect which characterizes the end-to-end performance of the transmission : frame transmission delay, delay variations and the proportion of lost or degraded blocks of image in a video flow for example. The conjunction of these two aspects defines the global Quality of Service at user level.

At the connection establishment time, QOS is negotiated with the network and with distant end system and a quantitative level is selected. Furthermore data are generated by application according to this level and the network is responsible of transporting these data while providing the necessary effort for maintaining the qualitative degree of performance negotiated.

Both QOS aspects may take different levels of value, according to the application and/or user. Some applications need very high levels of qualitative data delivery (zero loss for example) while they may (e.g. image processing) or not (e.g. assisted surgery) accept medium or low levels of quantitative performance. Other applications (adaptive applications), on the contrary, may deal with lower qualitative levels or temporary qualitative degradation : video transmission for example admit non zero loss (data restoration) and some flexibility in transmission delay and delay variation (by buffering).

We underline and this is the main assumption in this paper, that in most cases, qualitative aspect of QOS in multimedia data transmission is of higher importance to user(s) (and application) than the quantitative aspect. That means when QOS can not be maintained at the negotiated level and QOS degradation is unavoidable, it is more appropriate to guide the degradation so that only the quantitative level of the QOS is affected and the qualitative level could be maintained.

2.2. Resource requirements

Multimedia applications, specially those involving video transmission, requires more complex resources provisioning mechanisms than by past. Recent progress in multimedia data encoding standards (MPEG2 [2] for example) have substantially reduced the amount of data to be transported but as a result these flows become highly bursty, and sensitive (to loss and delay variation) while they remain resource demanding (mainly bandwidth, CPU time and memory).

To provide an application with some level of QOS, resources have to be allocated at sufficient quantity to the application and its associated connections. Then several strategies are possible, each one offering a different level of guarantee (guarantee

...