Pablo Belzarena

Early traffic classification using support vector machines

Internet traffic classification is an essential task for managing large networks. Network design, routing optimization, quality of service management, anomaly and intrusion detection tasks can be improved with a good knowledge of the traffic. Traditional classification methods based on transport port analysis have become inappropriate for modern applications. Payload based analysis using pattern searching have privacy concerns and are usually slow and expensive in computational cost. In recent years, traffic classification based on the statistical properties of flows has become a relevant topic. In this work we analyze the size of the firsts packets on both directions of a flow as a relevant statistical fingerprint. This fingerprint is enough for accurate traffic classification and so can be useful for early traffic identification in real time. This work proposes the use of a supervised machine learning clustering method for traffic classification based on Support Vector Machines. We compare our method accuracy with a more classical centroid based approach, obtaining promising results.

Quality of service parameters and link operating point estimation based on effective bandwidths

This work addresses the estimation and calculation of the operating point of a networks link in a digital traffic network. The notion of operating point comes from effective bandwidth (EB) theory. The results shown are valid for a wide range of traffic types. We show that, given a good EB estimator, the operating point, i.e. the values of time and space parameters in which the EB is related with the asymptotic overflow probability, can also be accurately estimated. This means that the operating point (and other parameters) inherits the statistical properties of the EB estimation. This affirmation is not an obvious one, because operating point parameters are related with the EB through an implicit function involving extremal conditions computations. Imposing some regularity conditions, a consistent estimator and confidence regions for the operating point and Quality of Service parameters are developed. These conditions are very general, and they are met by commonly used estimators as the averaging estimator presented in [C. Courcoubetis, R. Weber, Buffer overflow asymptotics for a switch handling many traffic sources, J. Appl. Probability 33 (1996)] or the Markov Fluid model estimator presented in [J. Pechiar, G. Perera, M. Simon, Effective bandwidth estimation and testing for Markov sources, Perform. Eval. 48 (2002) 157-175]. Using a software package developed by our group that estimates the EB and other relevant parameters from traffic traces, simulation results are compared with the analytical results, showing very good fitting.

Network Bandwidth Allocation via Distributed Auctions with Time Reservations

This paper studies the problem of allocating network capacity through periodic auctions. We impose the following conditions: fully distributed solutions over an arbitrary network topology, and the requirement that resources allocated in a given auction are reserved for the entire duration of the connection, not subject to future contention. Under these conditions, we study the problem of selling capacity to optimize revenue for the operator. We first study optimal revenue for a single distributed auction in a general network. Next, the periodic auctions case is considered for a single link, modelling the optimal revenue problem as a Markov decision process (MDP); we develop a sequence of receding horizon approximations to its solution. Combining the two approaches we formulate a receding horizon optimization of revenue over a general network topology, that yields a distributed implementation. The proposal is demonstrated through simulations.

Network bandwidth allocation with end-to-end QoS constraints and revenue sharing in multi-domain federations

Internet is evolving, traffic continues to grow, new revenue sources are sought by Network and Service Providers. Value added services with real time characteristics are likely to be common currency in the near future. Quality of Service (QoS) could allow Application/Service Providers (APs) to offer better services to the end users. At the same time, all actors claim for a fair distribution of revenues. Inspired by this scenario, we propose a complete framework for selling interdomain quality assured services, and subsequently distributing revenues, in an Autonomous System (AS) association context. We state the problem as a network utility maximization problem with QoS constraints and show that a distributed solution can be carried out. In order to fairly share the resulting revenue we study concepts from coalitional game theory and propose a solution based on the Shapley value and statistics on the revenues. Simulations of the whole proposal are shown.

Decision making in forward power markets with supply and demand uncertainty

The paper studies forward markets of electric energy, where generators and consumers bid for quantities of energy ahead of time, but face uncertainty on their real-time supply or demand, expressed through a probability distribution. The optimal forward decision is derived in both cases, providing extensions to the recent literature on this topic. In particular, the case where demand is elastic in addition to uncertain is addressed in detail. Finally, we analyze the integrated forward market where buyers and sellers of random energy interact with dispatchable sellers to determine a clearing price.

Optimal multipath forwarding in planned Wireless Mesh Networks

Wireless Mesh Networks (WMNs) have emerged in the last years as a cost-efficient alternative to traditional wired access networks. In the context of WMNs resources are intrinsically scarce, which has led to the proposal of dynamic routing in order to fully exploit the network capacity. We argue instead in favour of separating routing from forwarding (i.e. a la MPLS). Our proposal is a dynamic load-balancing scheme that forwards incoming packets along several pre-established paths in order to minimize a certain congestion function. We consider a particular but very typical scenario: a planned WMN where all links do not interfere with each other. We use a simple and versatile congestion function: the sum of the average queue length over all network nodes interfaces. We present a method to learn this function from measurements and several simulations to illustrate the framework, comparing our proposal with the IEEE 802.11s standard.

On the problem of revenue sharing in multi-domain federations

Autonomous System alliances or federations are envisioned to emerge in the near future as a means of selling end-to-end quality assured services through interdomain networks. This collaborative paradigm mainly responds to the ever increasing Internet traffic volumes that requires assured quality, and constitutes a new business opportunity for Network Service Providers (NSPs). However, current Internet business rules are not likely to satisfy all involved partners in this emerging scenario. How the revenue is shared among NSPs must be agreed in advance, and should enforce economical incentives to join an alliance and remain in it, so that the alliance remains stable. In this paper, we work on the scenario of such federations, where service selling is formulated as a Network Utility Maximization (NUM) problem. In this context, we formally formulate the properties the revenue sharing (RS) method should fulfill and argue why the existing methods are not suitable. Finally, we propose a family of solutions to the RS problem such that the economical stability and efficiency of the alliance in the long term is guaranteed. The proposed method is based on solving a series of Optimization Problems and considering statistics on the incomes.

End-to-end measurements over GPRS-EDGE networks

In the last years, QoS (Quality of Service) parameter estimation has become a main research area in networking due to a continuous growth of the Internet. End-to-end active measurement is one of the topics that focuses on this research area. However, these measurement methodologies have focused on end-to-end measurements over the wired Internet. The development of cellular data services is shifting the resarch focus on QoS from wired to wireless networks. End-to-end measurement methodologies of cellular networks have some issues that are not considered by traditional measurement techniques. This paper analyzes these issues and suggests an end-to-end active measurement methodology that deals with these particular problems. The proposed research and measurement methodology is based on a GSM (Global System for Mobile Communications) network, particularly on data services on a GPRS-EDGE (General Packet Radio Service/Enhanced Data Rates for GSM Evolution) network. Several experiments in different situations have been done in a real cellular network. These experiments have tested the performance of the methodology in different data access conditions.

Minimum queue length load-balancing in planned Wireless Mesh Networks

Wireless Mesh Networks (WMNS) have emerged in the last years as a cost-efficient alternative to traditional wired access networks. In order to fully exploit the intrinsically scarce resources WMNS possess, the use of dynamic routing has been proposed. We argue instead in favour of separating routing from forwarding (i.e. à la MPLS) and implementing a dynamic load-balancing scheme that forwards incoming packets along several pre-established paths in order to minimize a certain congestion function. In this paper, we consider a particular but very important scenario: a planned WMN where all bidirectional point-to-point links do not interfere with each other. Due to its versatility and simplicity, we use the sum over all links of the mean queue length as congestion function. A method to learn this function from measurements is presented, whereas simulations illustrate the framework.

End-to-end quality of service-based admission control using the fictitious network analysis

The performance analysis of a network link is a well-studied problem. However, the most interesting issue for a service provider is to evaluate the end-to-end quality of service (QoS). The evaluation of the end-to-end QoS (e.g. loss probability or delay) depends on the traffic statistic which is constantly modified as the traffic traverse the network, making its analysis a very difficult problem. In this work we use a simplified framework known as fictitious network analysis that allows us to estimate on-line the end-to-end loss ratio from input traffic traces statistics. We prove that the defined estimator is consistent and that a Central Limit Theorem is verified. Based on these estimations an admission control mechanism can be implemented. More precisely, we propose a simply method to estimate the control admission region, i.e. which are the flows that can be accepted in the network that verifies that its end-to-end loss ratio is smaller than a given threshold. While decisions based on the fictitious network analysis are safe, it may lead to network resources under-utilization (it generally overestimates the QoS parameters). In this work we establish sufficient conditions to assure that results obtained by means of the fictitious network coincide with real ones (there is no overestimation). We present first the conditions in the one-link case and extend them to the multilink case, necessary to evaluate the end-to-end loss ratio. When different results are obtained we define a method to find a bound for the overestimation. We also present numerical examples to compare the performance obtained in the real and the fictitious network, validating our main results.