A Measurement-Oriented Approach to Modeling Packet Loss in IP Networks

The support for quality of service has become an absolute must in present data communication networks, as they must now adapt in order to transport time-critical applications, such as voice and video. Great efforts have been made to make the best-effort IP infrastructure suitable for these next-generation applications. QoS frameworks such as IntServ and DiffServ are helping the goal of deploying virtually any communication over IP networks, with a preference of service providers to use DiffServ, due to a number of technological and operational reasons. However, DiffServ lacks complete formal specification-which negatively impacts performance and consistency across implementations-and many of the current partial DiffServ models are flawed by their lack of taking into account variabilities in packet sizes. As previous research has shown, this omission all but renders the models unusable in a number of scenarios. This paper proposes a measurement-based method to include such variabilities in mathematical models. The rationale for using a measurement-based approach versus an analytical one, as well as results and validation of this approach, are also discussed.