MediaNet

MediaNet is a distribution network for streaming data.  While the model is fairly general, our inspiration and current target has been to determine how to best deliver reconnaissance data (video and other relevant conditions) from a fleet of UAV's (Unmanned Aerial Vehicles) to a network of command and control centers.  The technical challenge is to deliver the data in resource-limited, small-scale (in terms of number of nodes, not necessarily distance between them) environments meeting the preferences and demands of the target control applications.

The MediaNet architecture is roughly as follows

• It uses a computational, overlay network to manage data delivery and user preferences.  User preferences are specified as graphs of computations, called continuous media networks, encoded in XML.  Each node in the graph transforms the incoming stream, and sends output to zero or more downstream nodes.
• To ensure good quality of service, a system-wide scheduler performs adaptive allocation of the nodes in the user preferences graph to nodes in the network; more than one user computation may be scheduled on a single network node.  In resource-limited situations, this allocation takes into account user preferences for adaptation.  In particular, the user will specify a series of graphs, all with a corresponding "utility" indicating the user's preference.

While MediaNet currently targets multimedia streams, our model is general enough to apply to more arbitrary distributed computations.  MediaNet compute nodes are written in Cyclone, a type-safe systems programming language.

People

• Michael Hicks
• Robbert Van Renesse

Papers and Software

User-specified Adaptive Scheduling in a Streaming Media Network.  Michael Hicks, Robbert van Renesse, Mark Bickford, Robert Constable, Christoph Kreitz, and Lori Lorigo.  Cornell Technical Report, June 2002. PS PDF

Software will be available on request.