Thèse soutenue

FR
Auteur / Autrice : Lucas Di Cioccio
Direction : Renata Teixeira
Type : Thèse de doctorat
Discipline(s) : Informatique
Date : Soutenance en 2013
Etablissement(s) : Paris 6

Résumé

FR  |  
EN

Broadband Internet access is now widespread and many users connect to theInternet from home. Often, Internet users at home are not computer. When aperformance problem occurs, users have no simple means to diagnose the problemand may call their Internet service provider to fix the problem, even if theproblem comes from the user network. This situation frustrates Internet usersand incurs a large cost on the Internet service providers which must provisioncall centers. In this thesis, we consider techniques for end-hosts to pinpoint whetherperformance problems occur in the home network or not. We show that some homenetwork configurations affect the end-to-end performance and that existingtechniques cannot always pinpoint whether the home network is the performancebottleneck. To get a better understanding of existing home networks at large,we design HomeNet Profiler, a software measurement tool to measure the list ofdevices active in the home network, the implementation of UPnP in residentialhome gateways, and the WiFi environment inside home networks. With our datasetconsisting of nearly 3000 homes, we show that home networks are often small butcan have up to 20 devices. We demonstrate that UPnP queries, can pinpointcross-traffic from the home network and differentiate local from wide-arealosses. We also show that the home WiFi environment is generally dense and hasan inherent risk for interference. To leverage and take advantage of this highWiFi density, we design neighbor-assisted diagnosis techniques. Thesetechniques are able to efficiently detect and distinguish uplink and downlinkdelays and loss rates with small error.