6/5/2008 -
Talk - Harsha Madhyastha
"An Internet-Wide Information Plane for Distributed Applications" at EBU3b (CSE) 4140 11:00 A.M.
Applications such as BitTorrent and Skype that are distributed over
millions of end-hosts across the Internet are becoming increasingly
common. These large-scale distributed applications need to optimize
their performance by carefully choosing which paths they use for
communication. However, the lack of performance guarantees from the
Internet makes it hard for applications to discover properties of
Internet paths. Instead of treating the Internet as a blackbox and
issuing only end-to-end measurements, I am going to argue in this
talk for a more structural approach for discovering path
information. I will demonstrate that it is feasible to uncover the
Internet's structure and use it to make predictions of path
properties at Internet-scale.
I will describe the design and implementation of iPlane, an
information plane we have built that can predict multiple metrics
such as latency, loss rate, and bandwidth capacity between any pair
of end-hosts on the Internet, even hosts that iPlane does not
control. iPlane's structural approach enables it to predict end-to-
end path performance by composing the performance of measured
segments of Internet paths. iPlane both hosts a query server that
can be queried via an XMLRPC interface, and distributes a compact
10MB annotated Internet map that end-hosts can download to make
local queries. We applied iPlane to several representative
distributed applications: content distribution, swarming peer-to-
peer filesharing, and voice-over-IP. In each case, using iPlane's
predictions resulted in improved application performance.
Bio:
Harsha V. Madhyastha is a PhD. Candidate at the University of
Washington. He received his B.Tech. from the Indian Institute of
Technology, Madras in 2003, and his M.S. from the University of
Washington in 2006. His research interests broadly encompass
building networked systems, and understanding the behavior and
properties of such systems.
Cooperative Association for Internet Data Analysis (CAIDA)