Tech Report CS-97-15

Broadcast Disks: Dissemination-based Data Management for Asymmetric Communication Environments

Swarup Acharya

September 1997

Abstract:

The increasing ability to interconnect computers through internetworking, wireless networks, high-bandwidth satellite, and cable networks has spawned a new class of information-centered applications based on {\em data dissemination}. These applications, which often serve huge client populations, employ broadcast to efficiently deliver data to the clients. In data dissemination, the data transfer is initiated by the server, inverting the traditional relationship between the client and the server. This thesis proposes a novel ``multi-disk'' framework for data dissemination called {\em Broadcast Disks}. The Broadcast Disks approach significantly improves upon the previous work in dissemination-based systems and raises a number of fundamentally new research challenges.

In this thesis, we first motivate why the rise of {\em asymmetric} environments (i.e., networks which have a significantly higher bandwidth available from servers to clients than in the reverse direction) and the scale of the emerging distributed information systems is causing a shift from the traditional {\em pull-based} client-server model to a {\em push-based} dissemination model. Then, the Broadcast Disks model is introduced and explored using a simulation-based study and a working implementation.

The bulk of this thesis uses simulation results to understand the basic tradeoffs in a dissemination framework. We demonstrate the performance benefits of the Broadcast Disks model over the traditional approach to structuring a dissemination program. We also introduce a new {\em cost-based} approach to client cache management and develop efficient algorithms for prefetching and dissemination of updates. Then, the thesis addresses the issues that arise in supporting clients using both push and pull-based data delivery. Finally, we describe our experience in building and testing a Broadcast Disks prototype. While the studies using the prototype validate the simulation-based intuitions, they also raise many new issues and highlight some of the shortcomings of the current technology for building push-based systems. The underlying theme driving the studies in this thesis is to develop techniques to improve system performance and scalability which adapt to the new tradeoffs in the emerging computing landscape.

(complete text in pdf or gzipped postscript)