Research Areas and Projects
This page describes the research areas and projects we are, or have been, involved in. These are: traffic characterization and modeling, network design, admission control and resource allocation, access networks, optical wireless communications and network security.
Traffic Characterization and Modeling
We research traffic models, and the associated parameter fitting procedures, capable of describing accurately the traffic characteristics with more impact on network performance.
We have developed parameter fitting procedures for both Markov Modulated Poisson Processes (MMPPs) and Batch Markovian Arrival Processes (BMAPs). MMPPs and BMAPs are attractive traffic models given the availability of mathematical tools for queuing analysis and for computation of effective bandwidths. The fitting procedures for MMPPs match the marginal distribution and the autocovariance of the arrival process. The fitting procedures for BMAPs match the packet size distribution, in addition to the marginal distribution and autocovariance of the arrival process. A major feature of these procedures is that the number of states of the MMPPs and BMAPs is not fixed a priori, and can be adapted to the particular trace being modeled.
We have introduced traffic models based on stochastic L-Systems and developed the associated fitting procedures. L-Systems are string rewriting techniques which were introduced by biologist A. Lindenmayer in 1968 as a method to model plant growth. They are characterized by an alphabet, an axiom and a set of production rules. The alphabet is a set of symbols; the production rules define transformations of symbols into strings of symbols; starting from an initial string (the axiom), an L-System constructs iteratively sequences of symbols replacing each symbol by the corresponding string according to the production rules. Traffic models based on stochastic L-Systems are very flexible in incorporating various types of traffic behavior (non-stationarity, LRD, self-similarity and multifractality) and were seen to achieve excellent fitting performance in terms of first and second order statistics and of queuing behavior (as assessed by the packet loss ratio and average delay).
We have analyzed Internet user traffic flows measured in a Portuguese ISP. Using singular-value decomposition, the weekly traffic utilization of each user was decomposed in periodic, noise and burst components. We have shown that this component can be used to obtain groups of users with similar weekly profiles of traffic utilization, without losing the main structure of the partition obtained from the complete data set that includes all three components.
We have studied the dependencies between Internet flow rates, flow sizes and flow durations considering the physical limitations imposed by the network and the experimental design, which could otherwhise lead to misleading results. These constraints were included in a parametric model, that we admit suitable to describe the relation between the logarithm of size and logarithm of duration of a flow. Theoretical results like moments, conditional distributions, conditional moments, and maximum likelihood estimators, among others, were obtained first for the reduced and than for the general parametric model.
We have developed a new approach based on neural networks for detection Internet applications. The NN will be trained using a set of known traffic values associated with each application and the corresponding aggregate traffic; after this training phase, the trained NN model can identify the traffic level associated with each application based on new values of aggregate traffic that are presented as inputs. The correlation that exists between the temporal sequence of aggregate traffic values and the current distribution of traffic per application is taken into account by presenting the current and the last h (where h represents a configurable parameter) values of aggregated traffic as inputs of the NN model.
We have specified, developed and evaluated a distributed traffic measurement system with a peer-to-peer architecture, named DTMS-P2P, which provide network managers with a tool to remotely configure third-party monitoring modules installed at different points of the network in order to perform test measurements. The system is based on a hierarchical overlay network organized in measurement areas. The overlay network is formed by two types of nodes, called probes and super-probes, which perform the measurements and store the measurement results. Super-probes have the specific role of providing connection among measurement areas and manage the exchange of messages between the network and the probes connected to them. The topology of the overlay network can change dynamically, with nodes being inserted and removed on-the-fly, and probes being transformed in super-probes and vice-versa, in response to changes in the available resources. The nodes collect two types of measured data: light data files and heavy data files. Light data files store the average round-trip time from each super-probe to every element (probes, super-probes and clients) it is connected to and are replicated in all super-probes, providing a coarse but widely available view of the network status. Heavy data files contain the results of detailed measurements carried out at the packet or flow level and can be replicated at some nodes of the overlay network. Replications are spread over the overlay network taking into account the resources available at nodes, so as to provide high resilience to failures. Users can configure traffic measurements and search the overlay network for measurement data through the so-called client element. The various tests carried out in the system have shown that it performs correctly and efficiently.
Actual IP networks must support a huge diversity of
applications and services and have to cope with many user behaviors and
different mechanisms of traffic generation and control. The combined effects of
all these factors lead to a highly variable traffic that brings increasing
challenges to network management operations. In this context, active traffic
monitoring is particularly important as it enables characterizing essential
aspects of network operation like, for example, quality of service measured in
terms of packet delays and losses. One-Way Active Measurement Protocol (OWAMP)
is a recent proposal of the Internet2 group that was standardized by IETF under
the scope of the IPPM group. OWAMP is an architecture used to perform active
measurements of one-way delays and losses between hosts. We have developed an
implementation in Java of OWAMP, named J-OWAMP, and carried out successfully
several interoperability tests with an implementation of the Internet2 group.
We proposed a new approach to the modeling of packet
switching networks, which is based only on measurements (or simulations) and
does not require any information on the network structure. Network managers
usually need to assess network behavior for varying number of users or evolving
traffic profiles, even without a complete knowledge of the network topology or
resources. In these cases, it is critical to have a framework that, based on
existing measurements and/or mathematical models (that describe the evolution of
the number of users and their traffic profiles), infer the future network
quality of service (QoS) parameters. Using these predictions, the network
manager can take substantiated decisions on the need for resources reallocation
or upgrading. The proposed network model is built from a set of (past)
measurements of the inbound/outbound traffic and of the corresponding Quality of
Service (QoS); the model can then be used to predict the (future) QoS based only
on the inbound/outbound traffic. There are no restrictions to the type of
metrics that characterize QoS. Our results, that include applying the procedure
to realistic network scenarios, show that this approach can achieve excellent
performance in terms of predicting the QoS of multiple network access links. The
QoS prediction is accurate even when there are significant changes in the number
of users.
In this area, we closely colaborate with Prof. António Pacheco and Prof. Rosário Oliveira, from the Department of Mathematics, Instituto Superior Técnico.
To find a list of our publications in this area click here.
People involved: Rui Valadas, Paulo Salvador, António Nogueira, Hélder Veiga, Eduardo Rocha.
We addressed the design of multiservice networks (e.g. ATM and MPLS). A main driving force of the research has been the formulation of linear integer programming models and the investigation of suitable solution techniques. Depending on the problems, solution techniques have been investigated in two domains: (i) through the search of heuristics (e.g., greedy based and Lagrangean relaxation algorithms) and meta-heuristics (e.g., genetic algorithms) and (ii) through valid reformulations that strength the linear relaxation of the dimensioning models. For the first class of solution techniques, suitable cut-sets that enable the calculation of good lower bounds have been investigated in order to access the quality of the solutions found. The following dimensioning problems have been addressed: (i) link capacity dimensioning to support explicit unicast routing, multicast routing, conversational and retrieval services, hierarchical logical networks, network survivability, and simultaneous consideration of flow level and packet level constraints, (ii) node dimensioning considering equipment costs relative to purchasing, location transfer and stocking and (iii) multi-layer dimensioning of MPLS over WDM networks. We have also developed software tools for the design of ATM and MPLS networks.
In this area, we closely colaborate with Prof. Luis Gouveia, from the Department of Statistics and Operational Research, Faculty of Sciences, University of Lisbon.
To find a list of our publications in this area click here.
People involved: Amaro de Sousa, Rui Valadas, Carlos Borges Lopes, Luis Cardoso, Jorge Patrão.
Admission Control and Resource Allocation
In large networks, individually managing each traffic flow on each of its traversed routers has fundamental scalability limitations, in both the control plane's requirements for signaling, state management, and admission control, and the data plane's requirements for per-flow scheduling mechanisms.
We have addressed the use probing mechanisms to assess the congestion state of the network and perform admission control at the end points. These mechanisms do not require any signaling protocol or any special packet processing within core nodes and, therefore, scale well for large networks. In particular, we proposed, studied and implemented a mechanism to overcome the so-called resource stealing problem that affects systems with probing-based admission control and multiple classes of service.
Aggregation is a mechanism used to reduce the number of signalling messages in an IntServ architecture. In this technique the admission control is only performed on an aggregated set of flows and therefore core routers need only to maintain the reservation state of each aggregate. A new version of the RSVP protocol is used in which a single RSVP reservation aggregates other RSVP reservations across a transit routing region. We have studied the tradeoffs between the signaling load and the resource utilization in network domains involving multiple aggregation regions.
We proposed and evaluated a new estimation methodology able to measure the capacity and the cross-traffic of all links belonging to the path between a source and a destination node. This approach resorts to the dispersion of packet pairs and trains sent sequentially in the network from the source to the destination. For this purpose, we make use of the concepts of ICMP timestamps and Traceroute. We then correlate the dispersions achieved for each packet pair and train in each sub-path from the source to the destination and measure the dispersion values in each network link. Finally, we estimate the capacity and cross-traffic in each link, which is done in previous work that only deal with the botleneck link.
In this area, we have collaborated closely with Prof. Ed Knightly, from Rice University.
To find a list of our publications in this area click here.
People involved: Rui Valadas, Susana Sargento.
In a joint work with Jorge Gonçalves and Teixeira de Sousa from PT Inovação we have proposed an architecture for IP-based Access Networks with QoS support. The architecture is targeted for the integrated support of broadband multimedia services, and designed to be low-cost and easily manageable. It includes SIP for session initiation, COPS and DIAMETER for QoS policy management and AAA and RSVP for resource reservation.
To find a list of our publications in this area click here.
People involved: Rui Valadas, Susana Sargento, Roger Salgado.
Optical Wireless Communications
Work in this area has been going on since 1991. Our main contributions have been: characterization of the noise provoked by ambient light, including the components due to fluorescent and incandescent lamps; characterization of the impulse response of the optical channel; analysis of modulation methods accounting for the combined effect of ambient noise and multipath dispersion; proposal and study of the utilization of angle diversity as a way to overcome ambient noise and multipath dispersion; proposal and study of solutions for interworking between wireless and wired local area networks; study of the impact of co-channel interference and optical power variations in the performance of medium access protocols. We have also participated actively in the specification of the IEEE 802.11 standard.
To find a list of our publications in this area click here.
Main people involved: Rui Valadas, Adriano Moreira, Cipriano Lomba, António Tavares, Rui Aguiar.
We are investigating several issues related with the security of TCP/IP networks, namely: (i) Intrusion Detection Systems (IDS), (ii) security in routing protocols, (iii) security in wireless LANs (IEEE 802.11), (iv) personal firewalls and personal IDSs, (v) IPSec and VPNs, (vi) tools for security auditing, vulnerability analysis, and log analysis, and (vii) anti-virus systems.
People involved: Rui Valadas, Cristiano Pereira, Hugo Marques, André Malheiro.