Traffic shaping based on an exponential token bucket for quantitative QoS: implementation and experiments on DiffServ routers

  • Authors:

  • E. Vayias;J. Soldatos;G. Kormentzas

  • Affiliations:

  • Intracom SA, 19.5km Markopoulou Ave, Paiania 19002, Greece;Athens Information Technology, 19.5km Markopoulou Ave, Paiania 19002, Greece;University of Aegean, Karlovassi, Samos 83200, Greece

  • Venue:
  • Computer Communications
  • Year:
  • 2006

Quantified Score

Hi-index 0.24

Visualization

Abstract

Support for quantitative QoS guarantees and associated Per-Hop Behaviours in a Differentiated Services architecture hinges on the specification of appropriate traffic control mechanisms. However, recent and past studies of Internet traffic have shown that it is rather hard to find a global and representative traffic model, which at the same time remains simple enough to facilitate the implementation of such mechanisms. In some previous works, we have demonstrated that suitable traffic shaping of packet flows that are subsequently multiplexed at a network node can permit the application of the M/G/1 model for performance modelling and control of the aggregate traffic stream. Using some handy control laws derived from the M/G/1 model, quantitative QoS guarantees can be provisioned end-to-end, across a series of multiplexers. In this paper we present our experiences from embedding this traffic-shaping scheme, which we call exponential token bucket filtering (etbf), in routers based on the Linux operating system. Initially, we describe how the traffic-shaping scheme can fit in a Differentiated Services architecture as a mechanism applied on the traffic transmitted from the edge routers to the core of a DiffServ domain, so that traffic multiplexed at the core routers conforms to the M/G/1 model. The Linux traffic control architecture is then presented, which enables the implementation of a DiffServ framework as well as how this shaping scheme is implemented within this architecture. Finally, results from experiments in a lab environment consisting of such Linux edge routers are presented. These results are in-line with findings from experiments and simulations in our previous works and show that the performance at the core multiplexing nodes can be controlled according to the M/G/1 model.