Tuesday, October 31, 2006

Paper: "iPlane: An Information Plane for Distributed Services"

iPlane: An Information Plane for Distributed Services
Harsha Madhyastha, Tomas Isdal, Michael Piatek, Colin Dixon, Thomas Anderson, and Arvind Krishnamurthy, University of Washington; Arun Venkataramani, University of Massachusetts Amherst


In this paper, we present the design, implementation, and evaluation of iPlane, a scalable service providing accurate predictions of Internet path performance for emerging overlay services. Unlike the more common black box latency prediction techniques in use today, iPlane adopts a structural approach and predicts end-to-end path performance by composing the performance of measured segments of Internet paths. For the paths we observed, this method allows us to accurately and efficiently predict latency, bandwidth, capacity and loss rates between arbitrary Internet hosts. We demonstrate the feasibility and utility of the iPlane service by applying it to several representative overlay services in use today: content distribution, swarming peer-to-peer filesharing, and voice-over-IP. In each case, using iPlane's predictions leads to improved overlay performance.


Blogger Shanth said...


Harsha Madhyastha presented iPlane, a
service that provides accurate
predictions of end-to-end Internet path
performance. He started off with the
observation that large-scale distributed
services, such as BitTorrent, depend on
information about the state of the
network for good performance. But most
current Internet measurement efforts,
such as GNP and Vivaldi, provide only
latency predictions between a pair of
nodes. In contrast, iPlane measures a
richer set of metrics, such as latency,
loss rate and available bandwidth.

iPlane continuously performs
measurements to generate and maintains
an atlas of the Internet by doing
traceroutes from a few distributed
vantage points. For scalability, targets
are clustered on the basis of BGP atoms
and a representative target from each
atom is used to approximate the
performance of targets in the atom.

iPlane uses structural information such
as the router-level topology and
autonomous system (AS) topology to
predict paths between arbitrary nodes in
the Internet. This prediction is made by
composing partial segments of known
Internet paths so as to exploit the
similarity of Internet routes. Next,
iPlane measures the link properties in
the Internet core and edge. In the
Internet core, the special vantage
points measure the link attributes. Link
properties at the Internet edges are
obtained by participating in BitTorrent
swarms and measuring the links to the
end-hosts while interacting with them.

Thus, to measure path properties between
any two hosts, first the path between
them is predicted. Then, iPlane composes
the measured properties of the
constituent path segments to predict the
performance of the composite path.

iPlane has been demonstrated to improve
the overlay performance of several
representative overlay services such as
content distribution networks, swarming
peer-to-peer filesharing and

During the Q&A session, Buck Krasic
(University of British Columbia)
observed that participating in
BitTorrent swarms to measure Internet
edge link properties might result in
conservative estimates, for e.g.
BitTorrent clients may have multiple
connections open and the bandwidth that
iPlane observes might just be a
fraction. He asked if iPlane had some
technique to compensate for that. Harsha
replied that they were using BitTorrent
to measure bandwidth capacity, not the
available bandwidth, and that bandwidth
capacity can be measured using a pair of
back-to-back packets.

1:29 PM  

Post a Comment

<< Home