Catálogo de publicaciones - libros

Wireless ad-hoc networks are based on shared medium technology where the nodes arrange access to the medium in a distributed way independent of their current traffic demand. This has the inherent drawback that a node that serves as a relay node for transmissions of multiple neighboring nodes is prone to become a performance “bottleneck”. In the present paper such a bottleneck node is modeled via an idealized fluid-flow queueing model in which the complex packet-level behavior () is represented by a small set of parameters. We extensively validate the model by ad-hoc network simulations that include all the details of the widely used 802.11 -protocol. Further we show that the overall flow transfer time of a multi-hop flow, which consists of the sum of the delays at the individual nodes, improves by granting a larger share of the medium capacity to the bottleneck node. Such alternative resource sharing strategies can be enforced in real systems by deploying the recently standardized -protocol. We propose a mapping between the parameter settings of and the fluid-flow model, and validate the fluid-flow model and the parameter mapping with detailed system simulations.

Given a placement of wireless nodes in space and a traffic demand between pairs of nodes, can these traffic demands be supported by the resulting network? A key issue for answering this question is wireless interference between neighbouring nodes including self interference along multi-hop paths. This paper presents a generic model for sustainable network load in a multi-hop wireless network under interference constraints, and recasts this model into a multicommodity flow problem with interference constraints. Using Farkas’ Lemma, we obtain a necessary and sufficient condition for feasibility of this multicommodity flow problem, leading to a tight upper bound on network throughput. Our results are illustrated by examples.

We present an economically efficient framework for provision of essential input for QoS-aware IP network planning. Firstly, we define a process for reuse of network accounting data for construction of a QoS-aware network demand matrix. Secondly, we define a process for estimation of QoS-related effective bandwidth coefficients from packet traces collected per traffic classe. Taken together, these processes provide the necessary input required to plan a network in accordance with QoS constraints. We present results of a sensitivity analysis of the demand estimation process, and of an economic analysis of the relative merit of deployment of our approach in comparison to a traditional direct measurement-based approach. We conclude that although there is a degree of inaccuracy in our network demand estimation process this inaccuracy is within acceptable bounds, and that this is offset by the potential for significant cost reductions for the ISP.

In future Internet, multi-network services correspond to a new paradigm that intelligence in network control is gradually moved to the edge of the network. As a consequence, the application itself can influence or determine the amount of consumed bandwidth. Thus the user behaviour may change dramatically. This impacts the Quality of Service (QoS) and the Quality of Experience (QoE), a subjective measure from the user perspective of the overall value of the provided service or application. A selfish user or application tries to maximize its own QoE rather than to optimize the network QoS, in contrast to a legacy altruistic user.

In this paper we present the IQX hypothesis which assumes an exponential functional relationship between QoE and QoS. This contribution is a first step towards the quantification of the QoE for edge-based applications, where an example of VoIP is taken into account. Starting from a measurement of the Skype application, we show the basic properties of selfish and altruistic user behaviour in accordance to edge-based intelligence. The QoE is quantified in terms of MOS in dependence of the packet loss of the end-to-end connection, whereby Skype’s iLBC voice codec is used exemplarily. It is shown that the IQX hypothesis is verified in this application scenario. Furthermore, selfish user behaviour with replicated sending of voice datagrams is investigated with respect to the obtained QoE of a single user. In addition, the impact of this user behaviour on congestion in the network is outlined by means of simulations.

We compare Forward Error Correction (FEC) and frame ACCumulation (ACC) to see which of the two schemes most effectively reduce frame loss rate for an aggregate of VoIP flows, sharing a network bottleneck. We model this bottleneck by a M/M/1/K queue and we analytically show that given certain assumptions, FEC is the best choice for low initial load at the bottleneck. Then, as the initial load increases, a crossing point is reached after which applying ACC is the better choice. We study this crossing point through numerical examples. Furthermore, we present numerical examples indicating that ACC is better than FEC in bandwidth limited network scenarios, while performance is more equal for packet processing limited scenarios, with FEC being the slightly better choice. Finally, we introduce more general queue models, e.g. the MMPP/M/1/K queue, to model traffic scenarios like the aggregate of ON/OFF VoIP traffic.

Cross-layer design has been proposed to optimize the performance of networks by exploiting the inter-relation among parameters and procedures at different levels of the protocol stack. This may be particularly beneficial in wireless scenarios, and for quality-of-service support. This paper proposes a quantitative study of cross-layer performance optimization for Voice over WiFi communications, which enables design engineers to analyze and quantify inter-layer dependencies and to identify the optimal operating point of the system, by using cost-benefit principles. Furthermore, insight gained on the problem enables the proposal of design principles for a Call Admission Control scheme able to enhance the overall system performance by limiting the number of users in the system and signalling to the active terminals of the proper parameter settings to optimize overall performance.

In this paper we aim to develop an analytical model of the cross-layer characteristics of a Near End Cross Talk (NEXT) limited ADSL channel and determine the effective channel capacity of such a system. The ADSL users often encounter impact of other users in the same cable system due to NEXT interference that reduces the transmission capacity of the ADSL link. At the same time, due to application behavior, the source traffic of the user introduces a level of uncertainty. This paper proposes a model for capturing the cross-layer dynamics of these two uncertainties of an ADSL system. The model provides a closed form analytical method for determining the effective capacity of the ADSL system limited by NEXT interference.

The issue of buffer sizing is rightly receiving increasing attention with the realization that the bandwidth delay product rule-of-thumb is becoming unsustainable as link capacity continues to grow. In the present paper we examine this issue from the light of our understanding of traffic characteristics and the performance of statistical bandwidth sharing. We demonstrate through simple analytical models coupled with the results of ns2 simulations that, while a buffer equivalent to the bandwidth delay product is certainly unnecessary, the recently advocated reduction to a few dozen packets is too drastic. The required buffer size depends significantly on the peak exogenous rate of multiplexed flows.

Flow level analysis of data networks has recently taken a major step towards tractability with the introduction of a resource sharing scheme called balanced fairness. We consider the balanced fairness concept in analyzing per-flow throughput in complex networks with a large number of flow classes. The two existing practical approaches in the setting, namely performance bounds and asymptotic analysis, require that the capacity set of the network is given explicitly as a set of (linear) constraints. We extend the asymptotic analysis method by providing explicit expressions for the second order throughput derivative in the light traffic regime. We show how asymptotic analysis can be applied in multipath routing and wireless networks, where the linear constraints cannot be readily worked out in explicit form. Finally, we introduce a numerical throughput analysis scheme based on Monte Carlo method.

In this paper, we propose a novel active queue management (AQM) algorithm called Neuron Control RED (NC-RED) that overcomes the drawbacks of the original RED. NC-RED uses a neuron controller to adaptively adjust the maximum drop probability to stabilize the average queue length around the target queue length. We demonstrate by simulations that NC-RED maintains stable operation independent of traffic loading, round trip propagation delay, and bottleneck capacity. We also demonstrate that NC-RED is robust to non-responsive UDP traffic and HTTP traffic, and it is effective for networks with multiple bottlenecks. Comparison with other well-known AQM algorithms like PI, REM and ARED demonstrates the superiority of NC-RED in achieving faster convergence to queue length target and smaller queue length jitter.