(A) methodology for internet-wide path and delay estimation from existing measurements

Abstract

Delay and path estimation between arbitrary points becomes even more and more crucial problem for distributed applications that heavily depend on network-internal performance. In order to deliver better performance to end-users, the Internet-wide services and applications depend on accurate information about the internal network state. Knowledge of network internals allow distributed applications to solve commonly encountered problems in their operations, such as nearest neighbor discovery, leader node selection, and optimal distribution tree organization. However, today`s Internet does not provide such information explicitly and application developers often resort to adhoc measurement tools to obtain the necessary data. This poses an additional tax on the development cost of new services and applications. In order to estimate or predict end-to-end delay accurately, a number of systems have been proposed and implemented. And these systems attempt to provide a shared measurement infrastructure for distributed applications. Though diverse in underlying mechanisms, these systems have the common goal of providing network-internal characteristics to applications, and their measurements overlap (or even wasted) significantly. Moreover, running these systems over a long period of time requires a significant effort. Not many of the proposed systems are running as of today. Longitudinal studies of the network performance are important to network operators and researchers and bring insights not available from one-time analysis. In view of the large variety of proposed (and already deployed) measurement systems, in this dissertation we opted for an alternative approach, measurement reuse, for the Internet path and delay performance estimation. We believe that it is about time that we look into the possibility of taking the most out of existing measurements and their infrastructures, instead of building yet another measurement system. Instead of performing measurements our...