Date of Award
Doctor of Philosophy (PhD)
Computer and Information Science
Boon Thau Loo
Components of networks, and by extension the internet can fail. It is, therefore, important to find the points of failure and resolve existing issues as quickly as possible. Resolution, however, takes time and its important to maintain high quality of service (QoS) for existing clients while it is in progress. In this work, our goal is to provide clients with means of avoiding failures if/when possible to maintain high QoS while enabling them to assist in the diagnosis process to speed up the time to recovery.
Fixing failures relies on first detecting that there is one and then identifying where it occurred so as to be able to remedy it. We take a two-step approach in our solution. First, we identify the entity (Client, Server, Network) responsible for the failure. Next, if a failure is identified as network related additional algorithms are triggered to detect the device responsible.
To achieve the first step, we revisit the question: how much can you infer about a failure using TCP statistics collected at one of the endpoints in a connection? Using an agent that captures TCP statistics at one of the end points we devise a classification algorithm that identifies the root cause of failures. Using insights derived from this classification algorithm we identify dominant TCP metrics that indicate where/why problems occur. If/when a failure is identified as a network related problem, the second step is triggered, where the algorithm uses additional information that is collected from ``failed'' connections to identify the device which resulted in the failure.
Failures are also disruptive to user's performance. Resolution may take time. Therefore, it is important to be able to shield clients from their effects as much as possible.
One option for avoiding problems resulting from failures is to rely on multiple paths (they are unlikely to go bad at the same time). The use of multiple paths involves both selecting paths (routing) and using them effectively. The second part of this thesis explores the efficacy of multipath communication in such situations.
It is expected that multi-path communications have monetary implications for the ISP's and content providers. Our solution, therefore, aims to minimize such costs to the content providers while significantly improving user performance.
Arzani, Behnaz Arzani, "Improving Network Performance Through Endpoint Diagnosis And Multipath Communications" (2017). Publicly Accessible Penn Dissertations. 2169.