Yatin Chawathe

Research Interests
Intel Research Seattle
AT&T Labs Research
Scattercast: An Architecture for Internet Broadcast Distribution as an Infrastructure Service

Yatin Chawathe

Despite the phenomenal success of the world wide web, one class of Internet applications that has yet to be satisfactorily realized is that of Internet broadcasting: the distribution of Internet content from one or more sources to a large number of simultaneous receivers. For large-scale broadcasting, the traditional Internet model of point-to-point unicast communication does not scale. So, the networking research community proposed IP multicast, a network layer service that allows a single source to distribute a data stream to many simultaneous receivers in an efficient manner. However, this network layer approach has met with limited success due to a number of factors including complexity of the network protocol itself, its inability to address Internet heterogeneity, and its lack of support for efficient and scalable transport protocols for reliability and congestion control. As a result, in spite of a decade of existence, the multicast protocol architecture remains just a research commodity with limited penetration into the commercially deployed Internet.

In this dissertation, we propose a new model for Internet broadcasting where we view multi-point delivery not as a network primitive but rather as an application-level infrastructure service. Our architecture relies on a collection of strategically placed network agents that collaborate with each other to form an overlay network composed of unicast interconnections. A data source distributes content to its receivers on top of this overlay network. We call this communication model scattercast and the network agents that are central to this model ScatterCast proXies or SCXs. The scattercast architecture effectively shifts the complexity associated with large-scale broadcasting away from the routing layer into a higher infrastructure service layer where it can be more easily managed. Moreover, by incorporating application-level intelligence within individual SCXs, scattercast can expand the traditional distribution model to optimize it for individual application needs by taking into account the effects of heterogeneity and application characteristics.

Such an infrastructural approach introduces two key questions: how do we construct the overlay network of SCXs in a distributed and dynamic fashion, and how do applications customize the overlay distribution framework to optimize it for their specific environment. To address the first problem, we present a protocol called Gossamer for grouping clients with SCXs and building an overlay network of unicast connections across SCXs, over which sources transmit data via efficient distribution trees. To address the second problem, scattercast develops a highly flexible and application-aware transport framework that incorporates the semantics of the application data into the transport protocol to allow the architecture to optimize the protocol for individual applications.

We demonstrate the efficacy of the scattercast approach using two complementary styles of evaluation. First, we use a set of simulations to demonstrate that the protocols and algorithms that underlie scattercast are viable and produce efficient overlay distribution networks. Second, we implement "real" applications on top of the architecture to demonstrate its ability to be customized for individual application requirements.

We believe that the scattercast approach of explicit application-level infrastructure embedded within the network is a new and promising direction for adaptive Internet applications such as network broadcasting. Similar approaches have recently been adopted by a number of commercial ventures for making the Internet broadcast-capable. In the future, we expect to see a greater proliferation of such infrastructural elements that enhance the Internet's capabilities while at the same time co-existing with the core IP architecture.

Back to Publications/Talks