|Ph.D Student||Roitman Haggai|
|Subject||Profile-Based Online Data Delivery: Model and Algorithms|
|Department||Department of Industrial Engineering and Management||Supervisor||Professor Avigdor Gal|
|Full Thesis text|
This dissertation studies online data delivery using profiles. Online data delivery requires to deliver data from servers to clients in a timely manner. Profiles specify client data delivery requirements as well as server capabilities to support different client requirements. Online data delivery is challenging in environments with a limited system resource budget, e.g., bandwidth constraints. We identify numerous applications to the methods developed in this work, including Web monitoring, Web feeds, data mashups, personalized services, and data caching. We propose a novel profile model for online data delivery and a flexible, simple, efficient, and scalable solution for delivery of data from servers (the data producers) to clients (the data consumers). The proposed profile model ProMo is unique in its ability to capture both client requirements and server capabilities using the same profile language. For this purpose we use an abstraction of execution intervals that serves as the basic building blocks of profiles. An execution interval associates a resource with a time period on which it is required by some client or can be delivered by some server. Combining execution intervals together can be used to specify both simple and complex data delivery profiles. Using the proposed profile model we address different aspects of data delivery and develop efficient frameworks and solutions for online data delivery. We first define the "proxy dilemma", formally addressing the tradeoff aspect of online data delivery. We provide a formal framework for data delivery tradeoffs based on Pareto optimality. We discuss both offline and online solutions to this problem. We then address the problem of satisfying complex data delivery needs in bandwidth-limited environments. We again discuss both offline and online solutions to this problem. We also demonstrate the importance of exploring the complexity level of client profiles and the distribution of the data access requests. Finally, we present the ProMo hybrid push-pull data delivery framework as a possible architecture for online data delivery. ProMo is a unique framework that utilizes the uniform structure of profiles. It can reason over server compatibilities for the aim of client profile satisfaction. We provide both a classification of server compatibilities and description of the resulting data delivery patterns. We then propose a hybrid push-pull solution that can be used to satisfy client profiles by maximizing the usage of server capabilities.