Date Awarded
2005
Document Type
Dissertation
Degree Name
Doctor of Philosophy (Ph.D.)
Department
Computer Science
Advisor
Weizhen Mao
Abstract
Inherent in the field of data broadcasting is a communication problem in which a server is to transmit a subset of data items in response to requests received from clients. The intent of the server is to optimize metrics quantifying the quality of service the system provides. This method of data dissemination has proved to be an efficient means of delivering information in asymmetric environments demanding massive scalability. of critical importance in such a system is the algorithm used by the server to construct a schedule of item broadcasts.;Due to the real-time nature of this problem, performances of heuristics designed to construct such schedules are heavily dependent on request instances. Thus it is challenging to establish the quality of one algorithm over another. Though several scheduling methods have been developed, these algorithms have been studied with a reliance on probabilistic assumptions and little emphasis on analytical results.;In contrast, we provide a formal treatment of the data broadcast scheduling problem in which analytical methods are applied, complemented by simulation experiments. Utilizing a worst-case technique known as competitive analysis, we establish bounds on the performance of various algorithms in the context of several different broadcast models. We describe results in three different settings.;Minimizing the total wait time of all requests with a single channel and multiple database items we establish the competitive ratios for two well-known algorithms, First Come First Served (FCFS) and Most Requests First (MRF) to be equal, and provide a general lower bound for all algorithms in this context. We describe simulation results that indicate the superior performance of MRF over FCFS on average. Minimizing two conflicting metrics, the total wait time and total broadcast cost, with a single channel and single database item we develop two on-line algorithms, establish their competitive ratios, and provide an optimal off-line algorithm used to simulate the impact of various parameters on the performance of both on-line heuristics. Finally, we extend the previous model by including multiple database items and establish a lower bound to a greedy algorithm for this context.
DOI
https://dx.doi.org/doi:10.21220/s2-0er6-b181
Rights
© The Author
Recommended Citation
Hawkins, Aaron Thomas, "Data broadcast scheduling: Models, algorithms, and analysis" (2005). Dissertations, Theses, and Masters Projects. William & Mary. Paper 1539623465.
https://dx.doi.org/doi:10.21220/s2-0er6-b181
