Date Awarded


Document Type


Degree Name

Doctor of Philosophy (Ph.D.)


Computer Science


Bruce B Lowekamp


Monitoring end-to-end available bandwidth is critical in helping applications and users efficiently use network resources. Because the performance of distributed systems is intrinsically linked to the performance of the network, applications that have knowledge of the available bandwidth can adapt to changing network conditions and optimize their performance. A well-designed available bandwidth tool should be easily deployable and non-intrusive. While several tools have been created to actively measure the end-to-end available bandwidth of a network path, they require instrumentation at both ends of the path, and the traffic injected by these tools may affect the performance of other applications on the path.;We propose a new passive monitoring system that accurately measures available bandwidth by applying self-induced congestion analysis to traces of application-generated traffic. The Watching Resources from the Edge of the Network (Wren) system transparently provides available bandwidth information to applications without having to modify the applications to make the measurements and with negligible impact on the performance of applications. Wren produces a series of real-time available bandwidth measurements that can be used by applications to adapt their runtime behavior to optimize performance or that can be sent to a central monitoring system for use by other or future applications.;Most active bandwidth tools rely on adjustments to the sending rate of packets to infer the available bandwidth. The major obstacle with using passive kernel-level traces of TCP traffic is that we have no control over the traffic pattern. We demonstrate that there is enough natural variability in the sending rates of TCP traffic that techniques used by active tools can be applied to traces of application-generated traffic to yield accurate available bandwidth measurements.;Wren uses kernel-level instrumentation to collect traces of application traffic and analyzes the traces in the user-level to achieve the necessary accuracy and avoid intrusiveness. We introduce new passive bandwidth algorithms based on the principles of the active tools to measure available bandwidth, investigate the effectiveness of these new algorithms, implement a real-time system capable of efficiently monitoring available bandwidth, and demonstrate that applications can use Wren measurements to adapt their runtime decisions.



© The Author