In this paper we report on a study of possible service level agreements in an IP network employing differentiated services. We discuss the nature of the quality of service guarantees given to network flows and relate this to the capacity provisioning processes of network operators.
A contribution of this paper is to address the way service level agreements might be determined from a coherent collection of models of network phenomena which themselves naturally operate on widely differing time–scales. The very fastest timescales within IP routers. The next fastest time-scale relates to session level controls embedded within the end-system behaviour of the TCP/IP congestion avoidance algorithms, operating in the range of milliseconds to seconds. The per-packet routing and the management of aggregated traffic flows can take place over time‐scales ranging from seconds to minutes to days. Provisioning of network resources takes place over intervals of weeks and months. All of these phenomena influence the overall structure of service level agreements.
This paper highlights the use of quantitative modelling methods which address fundamental concerns for network operators seeking to provide differentiated IP Quality of Service. The work described here is at a preliminary stage, but provides strong motivation for both further study and experimental validation. Our tentative conclusion is that the DiffServ Quality of Service mechanism is unlikely to be able to provide real measurable distinctions between classes on a pure IP network with no access restrictions, without either bandwidth partitioning at a lower layer or gratuitously damaging some traffic. It will, however, function as a back-stop minimum guaranteed level in times of congestion.