Description
Date depot: 1 janvier 1900
Titre: Foundations of an Internet measurement science
Directeur de thèse:
Dario ROSSI (LTCI (EDMH))
Domaine scientifique: Sciences et technologies de l'information et de la communication
Thématique CNRS : Non defini
Resumé:
{{Abstract}}
Internet measurements are often considered as a non scientific discipline, though, implicitly, this is already the case since the so called “passive techniques” (where Internet traffic is merely observed) remind the induction and deduction methods of Aristotle, while “active techniques” (where a reaction to a perturbation actively induced on the Internet is the observable of interest) are reminiscent of the experimental method of Galileo Galilei.
Yet, a major drawback currently affecting active experiments is their lack of reproducibility, which threatens their relevance and reliability; conversely, passive techniques are often applied to few measurement points and for a short duration, which limits the significance and scope of the deductions. The goal of this thesis is to introduce a rigorous scientific method in Internet measurements, by continuously repeat, integrate and correlate data from different Internet ``sensors’’, and to systematically repeat the analysis that are now done sporadically at best
{{PhD Thesis description}}
The ubiquity of Internet access, and the wide variety of Internet-enabled devices and applications, have made the Internet a principal pillar of the Information Society. Decentralized and diverse, the Internet is resilient and universal. However, its distributed nature leads to operational brittleness and difficulty in identifying and tracking the root causes of performance and availability issues. The first step to improve this situation is Internet measurement: illuminating the currently obscure dynamics of the Internet.
Generally speaking, networks have been built conceptually around the pillars of data and control planes, respectively to move data around and to instruct the network on how to precisely move this data. However, Internet architects had not foreseen its growth, and have therefore neglected the need for yet another crucial plane: the measurement plane. The role of this (non yet-existing) plane would be observe, measure and quantify the behavior of the network. Such actions are not only necessary to ensure the proper working of control and data planes, but also and especially to pinpoint the occurrence of performance problems, and possibly assist in determining their root causes. Given that such a measurement plane has not been conceived from the Internet infancy, several patched solutions for the network Operation & Management (OAM) have been added to the initial architectures over the years.
However, while several OAM tools exist, the lack of coordination of these OAM tool, the dynamic and changing nature of the Internet, its distributed nature and its very large scale, the fact that problem can happen at several layers, and on organization with different administrative boundaries make Internet OAM a very though job.
This thesis will provide significant advances with respect to the state of the art in measurement tools and methodology. Clearly, we are not the first to observe the above problems: hence, providing an exhaustive list of tools that have been engineered in the past would be needlessly verbose (see http://www.icir.org/models/tools.html or http://www.slac.stanford.edu/xorg/nmtf/nmtf-tools.html). Here we instead focus on some of the most recent and interesting developments in the area. Measurement tools can be roughly divided into passive, where the tool captures traffic as it goes by, or active, where the tool artificially injects traffic into the network in order to perform measurements. Perhaps the most commonplace passive tools are tcpdump or Wireshark[T1] and NetFlow[T2] designed to let researchers analyse the captured packets or flows, respectively. Beyond these, there are a large number of tools that designed to perform passive measurements, e.g., PF_RING[T3], Tstat[T4] and the CoralReef suite[T5]. Some tools explicitly target specific scenarios (the Aqualab project[T6] to monitor BitTorrent performance to check for changes that might indicate problems with the network), or focus on detecting network neutrality violations (Web Tripwires[T7] detect changes to web pages at some intermediary node; while NANO[T8], Netalyzr[T9] and Glasnost[T10] analyse various properties of an Internet connection, including blocking or shaping of specific services).
An equally large number of active measurement tools exist. Several of these have the aim of mapping the structure of the network, attempting to discover the topology of the Internet at a certain level in the protocol stack, usually IP or above. Operators use the results of such measurements in routing and network path optimization. The most obvious example in this area is the ubiquitous traceroute utility, which attempts to trace all IP hops between a source and a destination. iPlane[T11] is a service providing Internet path performance predictions by building an annotated map of the Internet, with measurement points deployed on PlanetLab. Merlin[T12]
Doctorant.e: Gong Yixi