Abstract

This work focuses on quality of service in large scale packet switching networks, like the Internet. An integrated services architecture has been defined by the IETF. It uses the RSVP protocol and is thus named IntServ/RSVP. However, some problems remain, especially considering interconnection networks, namely difficulties related to high rates and to the large number of flows to be managed.

During the last few years, most of high speed network architectures have evolved toward ``IP switching'' solutions. They are able to associate IP's robustness and flexibility with ATM switching engines' high capacity. This approach has been generalised as ``label switching''. Our first contribution is the ``RSVP switching'' proposal, a solution that integrates the IntServ/RSVP architecture in an ``IP switching'' environment. Even if it's able to combine quality of service management and high rates, this solution suffers from scalability problems due to the per-flow paradigm used by IntServ/RSVP.

Our second contribution is the study of aggregation techniques as means of providing quality of service without requiring management of per flow state. We show how managing per aggregate flows can be sufficient and we propose an architecture implementing such techniques in the ``aggregation zone''. We also describe mechanisms to interface transparently IntServ/RSVP with the aggregation based architecture.

Aggregation is a change in granularity that implies losing some flexibility in resources management. In our third contribution, we study a different approach allowing flow level granularity without requiring per flow state management. This is achieved by conditioning traffic at ``stateless zone'' entry, and limiting network nodes' impact on traffic characteristics.

Key Words:

QoS, quality of service, scalability, high rates, aggregation, IP switching, weighted fair queuing, virtual clock