Christof Brandauer

Stability of RED with two-way TCP traffic

This paper points out a phase effect with RED-like queue management algorithms and two-way TCP traffic. It is shown by simulation and measurement that the RED (random early detection) queue converges to a state of heavy oscillation in the presence of two-way bulk-data TCP traffic, causing suboptimal link utilization and service differentiation between in-profile and out-of-profile packets. Besides investigating the reasons for the oscillatory behavior, we show that the oscillations persist in scenarios with two-way Web-like TCP traffic and disappear in the case of significant amounts of cross traffic and/or usage of drop-tail instead of RED.

A Quantitative Model for the Parameter Setting of RED with TCP Traffic

This paper systematically derives a quantitative model how to set the parameters of the RED queue management algorithm as a function of the scenario parameters bottleneck bandwidth, round-trip-time, and number of TCP flows. It is shown that proper setting of RED parameters is a necessary condition for stability, i.e. to ensure convergence of the queue size to a desired equilibrium state and to limit oscillation around this equilibrium. The model provides the correct parameter settings, as illustrated by simulations and measurements with FTP and Web-like TCP flows in scenarios with homogeneous and heterogeneous round trip times.

AC algorithms in AQUILA QoS IP network

This paper presents the admission control (AC) algorithms that are implemented in the prototype multi-service AQUILA QoS IP network [4]. The discussed algorithms were developed for regulating traffic submitted to network services (NS) dedicated for handling: (1) real-time streaming traffic of variable bit rate (VBR) type and (2) elastic traffic, produced by greedy TCP sources. For the former NS, named Premium VBR, the measurement-based AC (MBAC) scheme is proposed. The applied approach employs the well-known Hoeffding bound formula [2] for calculating required link capacity to assure packet loss ratio at a predefined level. Moreover, this method is supported by the declarations about the peak bit rate as well as by the measurements of mean bit rate on aggregate flow level. For the latter NS, named premium multimedia, two alternative AC algorithms are investigated, both designed for assuring requested TCP throughput. While the first of them follows the token bucket marking (TBM) concept [12], the second one adjusts the advertised TCP window size to enable an ideal TCP behaviour (i.e. lossless packet transfer). The simulation results are included for illustrating the advantages of the discussed algorithms.

MINER - A measurement infrastructure for network research

Network measurements often require the coordinated use of multiple tools on distributed vantage points. Without any form of supporting command and control system, the management and execution of a systematic measurement study is a tedious mission. In this paper we present MINER which seeks to simplify this task. MINER is a programmable measurement infrastructure that integrates existing measurement tools and provides its users higher-level services on top of them. It enables users to define measurement activities, schedule executions and retrieve their results. The services are accessible via a tool-agnostic and unified programming interface with which measurement applications can be developed. We discuss MINERs requirements and design considerations and present a modular implementation that can be flexibly extended through plugins. Finally, we report on usage scenarios and directions for future work.

An implementation of a service class providing assured TCP rates within the AQUILA framework

This paper investigates an attempt to establish a QoS class that supports long-lived, bulk-data TCP flows that require a minimum rate from the network. The approach is based on a model for TCP flows subject to token bucket marking at the network edge and preferential dropping in the core network. The service class adds admission control functionality and a model for multi-RED queue management to the token bucket marker. The difficulty of parameterizing the mechanisms is discussed and analyzed in an explorative simulation study. A set of configuration parameters that enables a successful operation of the service class is identified and the achievable service provisioning is shown.

A Rate Controller for Long-Lived TCP Flows

In this paper a new mechanism for providing an assured rate to a long-lived TCP flow is proposed. The mechanism is called TCP rate controller (TRC) and operates as a traffic conditioner at the edge of a network. The TRC seeks to achieve the requested rate by imposing well directed drops and (artificial) delays on the flows packets. The choice of drop probability and delay is based on an analytical model of TCP sending behavior. It is shown in a simulation study that the TRC performs well over a broad range of requested rates and network RTTs.

Towards a Communication Network Model Generator for Evaluating Smart Grid Applications

In the past, various communication infrastructures and networks, network technologies and topologies, and protocols have been discussed to intelligently interconnect nodes across electric power distribution networks forming a smart grid. Similarly, attempts to esteem bandwidth requirements of smart grid applications under different assumptions were made, by applying models such as the queuing theory. The present paper aims at presenting a conceptual model allowing for the generation and simulation of characteristic communication networks, as part of the communication infrastructure of electric power distribution grids. Feasible input parameters are discussed and fields, where simulation can be used, are presented.