Streaming Media Multicast Delivery TechniquesMulticast delivery schedules for on-demand and near-live delivery of streaming media (e.g., audio and video). The schedules explicitly minimize the server disk and network bandwidth required to serve a set of clients, and so indirectly reduce network load.
Bandwidth Skimming: A Technique for Cost-Effective Video-on-Demand , Derek Eager, Mary Vernon, and John Zahorjan, Multimedia Computing and Networking 2000 (MMCN00), San Jose, CA, Jan. 25-27, 2000.
Optimal and Efficient Merging Schedules for Video-on-Demand Servers, Derek Eager, Mary Vernon, and John Zahorjan, Proc. 7th ACM Multimedia Conf. (Multimedia '99), Orlando, FL, Oct. 30 - Nov. 5, 1999.
Minimizing Bandwidth Requirements for On-Demand Data Delivery, Derek Eager, Mary Vernon, and John Zahorjan, Proc. 5th Int'l. Workshop on Multimedia Information Systems (MIS '99), Indian Wells, CA, October 21-23, 1999. (Award paper; forwarded to a journal.)
|Distributed 3D Real-time Rendering at Washington|
|(with Thu Nguyen)|
The goal of this work is to investigate system policies to support distributed, real-time 3D rendering. The work evolves from our previous experience in parallel and distributed system scheduling. In contrast to that work, though, the application(s) to be supported are real-time (rather than the long running, scientific applications typically studied in this area), with a VRML browser being the focus of our prototype implementation. Additionally, in contrast to most existing work on distributed rendering, we employ the graphics acceleration hardware on multiple nodes, as well as CPU, memory, and network bandwidth.
Image Layer Decomposition for Distributed Rendering on NOWs, Thu Nguyen and John Zahorjan, Proc. 2000 International Parallel and Distributed Processing Symposium, Cancun, Mexico, May 1-5, 2000.
Abstract from NSF Proposal
Scheduling Policies to Support Distributed 3D Multimedia Applications, Thu D. Nguyen and John Zahorjan. Proc. Sigmetrics/Performance '98, June 1998. (An extended version is available as Technical Report UW-CSE-97-11-03, University of Washington. [PDF, 321KB])
Processor and memory scheduling policy development and evaluation for both shared-memory and message passing parallel processors.
Completed work evaluates the effectiveness of array restructuring to improve locality. Array restructuring chooses a layout for multi-dimensional arrays in physical memory that is intended to match the access pattern exhibited by the code. It is a natural complement to control restructuring (e.g., tiling), with the advantage that array restructuring can be applied even when a dependence analysis cannot be performed.
Current work focuses on improving the execution time of sequential portions of a parallel program through the use of software controlled prefetching.
(with Shun-tak Leung)
How to efficiently determine a parallel schedule at execution time for loops that cannot be parallelized statically.
Run-Time Support for Space-Based Applications
(with Immaneni Ashok)
Space-based applications are typified by the "particles in space" problem, in which the goal is to simulate the effect of mutual forces on a set of particles operating in a 2D or 3D space. Our work developed both techniques to support development of parallel programs for this class of application and a prototype run-time system implementing them.
Support for Mobile Applications
(with George Forman)
Runtime support for applications that access global resources and must deal with variable levels of available service. A specific example is mobile applications, which may experience variable and unpredictable bandwidth.