Showing papers by "Andrzej Chydzinski published in 2009"

On the Evaluation of the Active Queue Management Mechanisms

Abstract: The active queue management in Internet routers exploits the idea that an incoming packet can be dropped by the router even if there is some available buffering space in the router. Therefore, the router can send congestion signals to TCP before the actual congestion occurs, preventing queues and delays from growing too high. Many active queue management algorithms has been proposed till now. Unfortunately, it is difficult to compare their performance because the authors often describe new features of a proposed algorithm using some specific network scenarios. In fact, among hundreds of papers it is hard to find a few that use exactly the same network scenario for evaluation of the proposed algorithm. In this paper, we propose a common testbed for the evaluation of active queue management mechanisms. It includes a specification of the network topology, link bandwidths and delays, traffic patterns on the application level, transport protocols, congestion level, bidirectional traffic characterization, metrics for the performance evaluation etc. As all active queue management algorithms cooperate closely with TCP, their evaluation methodology has to be closely connected with the TCP evaluation methodology. Therefore the testbed presented herein is based on the newest TCP evaluation suite (by Lachlan Andrew et al.), properly tailored for the active queue management evaluation purposes.

On the Performance of AQM Algorithms with Small Buffers

Abstract: Packet buffers in routers play a major role in congestion control in existing Internet. They compensate for the incoming traffic bursts transmitted by aggressive TCP applications. The appropriate sizing of buffers is important for providing equilibrium between the high link utilization, the loss ratio and the queueing delay. There is a growing need for routers with small buffers in high speed networks. In this paper we investigate the performance of active queue management algorithms with very small buffers. In particular, we show, that the use of very small buffers does not influence throughput on bottlenecked links if mixed (TCP / UDP) traffic is involved. On the other hand a very small buffer may lead to degradation of the inter-flow fairness.

On the deployment of AQM algorithms in the internet

Abstract: The active queue management algorithms for packet queueing in Internet routers have proven to be beneficial in terms of queue size and delay reduction. Before the wide deployment of the AQM in the Internet it should be checked first if it is beneficial to implement an AQM algorithm on only one router in a large network, composed of many bottlenecks and routers. It is also important to check whether an AQM algorithm implemented on several routers simultaneously can lead to undesired performance degradation, through mutual interactions of several AQM routers. In this paper we try to answer these questions using the newest traffic models designed to AQM testing and a network topology that consists of four routers and three bottleneck links. Comparing the performance of five queue management algorithms in three cases: (a) no AQM in the network, (b) AQM implemented on one router only, (c) all routers use the same AQM, we show that usage of AQM improves the average network performance, no matter if AQM is implemented on one router or on all routers. However, in some specific scenarios implementing an AQM algorithm on one router may lead to performance degradation on other, non-AQM routers.

On the stability of AQM algorithms

Abstract: Stable and predictable queue size is one of the main goals of the active queue management in Internet routers. A stable queue enables to maintain high throughput and low delay jitter and therefore has a positive impact on several Internet applications. It is not easy to design an AQM algorithm that would be able to maintain a stable queue size in every network conditions. For instance, it is known that large propagation delays have a negative impact on the queue size stability. In this paper, we demonstrate other factors that are likely to make the queue size unstable. Using the newest AQM testbed we show that, in addition to large delays, a low congestion level and low target queue size make stabilizing the queue size difficult and even the best known algorithms do not work well under these conditions.

A Program Suite for Distributed Simulation of Networks in Multi-core Environments

Abstract: The ns-2 simulator is one of the most popular and widely used discrete event network simulators. Its usage of the general purpose Tcl scripting language accompanied with a wide range of networking technologies allows users to easily create complex and differentiated simulation scenarios. The main limitation of ns-2 is its lack of support for multiprocessing, especially considering that the multi-core environments have recently become tools of everyday use. In this paper we (a) propose a programming approach that allows running single-model simulations simultaneously on a multi-core systems using ns-2 and (b) present a set of software tools developed following the proposed convention. In addition, we describe a lightweight software scheduler extending the use of ns-2 on ad-hoc grid systems.

Interactions between AQM routers in the Internet

Abstract: In this paper we present a study on mutual interactions between queue management algorithms implemented in Internet routers. The aim of this research was to detect possible issues connected with deployment of the active queue management (AQM) algorithms in the Internet. For that purpose, we simulated a net- work with four routers and checked its performance in different configurations of queue management algorithms. In particular, three basic configurations were considered: all routers in the network use the passive queue management, only one router in the network uses the active queue management (other routers use the passive queue management), all routers in the network use the active queue management. Six different AQM algorithms were used and tested in two congestion scenarios. In every case the performance of all bottleneck links was observed as well as the average performance of the whole network. In all the simulations, the average performance of the network, measured in terms of the queueing delay, was improved by when at least one router used an AQM algorithm. Moreover, in most cases the performance of every bottleneck link was also improved. These results give strong arguments for using AQM in the Internet. However, in some experiments we observed a performance degradation of particular bottleneck links when an AQM algorithm was used, indicating that some problems may indeed arise due to the mutual interactions of AQM routers.