Distributed Optimization for Information Dissemination

Overview

Recent years have witnessed the emergence of distributed information dissemination applications such as stock tickers, program trading, medical alerting, environmental monitoring, airline ticket pricing. These applications have a heterogenous set of requirements, such as frequent data transmissions, low end-to-end communication latency, large number of participants. They are generally deployed on fixed overlay topologies, sometimes several applications sharing the same infrastructure, and disseminate data using predetermined paths. Our project focuses on how to improve end-to-end data dissemination through efficient allocation of resources on the path (network bandwidth and CPU). We use the concept of utility function to measure the benefit of an application and formulate the resource allocation as a constraint optimization problem. We develop two distributed algorithms, LRGP and LLA, that optimize the overall application utility while keeping the resources of the infrastructure uncongested.

People

Cristian Lumezanu
Sumeer Bhola
Mark Astley

Papers

Online Optimization for Latency Assignment in Distributed Real-Time Systems
Cristian Lumezanu, Sumeer Bhola and Mark Astley
ICDCS, 2008
[pdf]

Utility Optimization for Event-Driven Distributed Infrastructures
Cristian Lumezanu, Sumeer Bhola, and Mark Astley
ICDCS, 2006
[pdf] [slides]