Detailed and understandable network diagnosis
Ratul Mahajan
Microsoft Research
Friday, November 06, 2009
1:30 PM, 2310 CS
By focusing on small enterprise networks, we discover that the state of art in network diagnosis lacks two key properties. The first property is detail -- the ability to diagnose application-specific faults (e.g., error codes) and to identify culprits at a fine granularity (e.g., a process or firewall configuration). We build a system, called NetMedic, that enables detailed diagnosis with minimal application knowledge. It is based on an intuitive technique that leverages the joint behavior of two interacting components (e.g., processes) in the past to estimate the likelihood of them impacting one another in the present. Our evaluation shows that for faults that we inject in a live environment, NetMedic correctly identifies the faulty component as the most likely culprit in 80% of the cases.
The second property is understandability -- the ability of system administrators to verify the output of a diagnostic system. The need for this verification arises because diagnostic systems that aim to diagnose a wide range of faults are not guaranteed to be always correct. To aid understandability, we build NetClinic, a novel visual analytics system. NetClinic enables administrators to verify the correctness (or incorrectness) of the automatic analysis at different levels of abstractions and allows them to move seamlessly across levels while retaining appropriate context. Our user study finds that users are able to accurately identify the culprit whether or not the diagnostic system suggested the right culprit.
Bio: Ratul Mahajan is a Researcher at Microsoft Research. His research interests include all aspects of networked systems, especially their architecture and design. His work spans Internet routing and measurements, incentive-compatible protocol design, practical models for wireless networks, and vehicular networks. He has published over 25 papers in top-tier venues such as SIGCOMM, SOSP, and NSDI. He is a winner of the SIGCOMM best paper award, the William R. Bennett Prize, and Microsoft Research Graduate Fellowship. He obtained his Ph.D. from the University of Washington (2005) and B.Tech. from Indian Institute of Technology, Delhi (1999).
Additional CS events.
|
