Reliability and timeliness analysis of content-based publish/subscribe systems

Publish/Subscribe;Analytical model;Performance modeling;Reliability;Timeliness

Pongthawornkamol, Thadpong

关键词: Publish/Subscribe;    Analytical model;    Performance modeling;    Reliability;    Timeliness;   

Others  :  https://www.ideals.illinois.edu/bitstream/handle/2142/18548/Pongthawornkamol_Thadpong.pdf?sequence=1&isAllowed=y

美国|英语

来源: The Illinois Digital Environment for Access to Learning and Scholarship

【 摘 要 】

Content-based Publish/subscribe systems (CBPS) is a simple yet powerful communication paradigm. Its content-centric nature is suitable for a wide spectrum of today's content-centric applications such as stock market quote exchange, remote monitoring and surveillance, RSS news feed, and online gaming. As the trend shows that the amount of information along with its producers become astonishingly increasing everyday, a publish/subscribe system seems to be one of only a few viable choices that could govern the next-generation world of communication. However, the content-centric nature of a publish/subscribe system also poses difficulty in analyzing or assessing its performance. Moreover, the complexity increases when deploying a publish/subscribe system on top of best-effort, unreliable wide-area networks. Such uncertainty and complexity become a hindrance to apply content-based publish/subscribe systems to delay-sensitive applications that require reliable/timely event delivery and tight resource control such as soft real-time systems or cyber-physical systems. The need to solve such problem calls for a good analytical model that could capture both expressiveness and uncertainty nature of distributed CBPS systems yet predict the system behavior accurately.This dissertation is, to the best our knowledge, the first attempt to analyze the reliability/timeliness performance of distributed content-based publish/subscribe systems under best-effort networks. It proposes a probabilistic, analytical framework of content-based publish/subscribe systems under different dynamism for the purpose of performance analysis. Specifically, given a publish/subscribe system configuration and dynamism parameters, it estimates event delivery probability and timeliness received by each subscriber in the publish/subscribe system. The dissertation also presents evaluation results of the proposed predictive model via simulations with both synthetic traces and real-world traces. The results yield prediction accuracy and effectiveness of the proposed framework. The proposed analytical framework can be used as a tool for performance assessment or as a building block for publish/subscribe system optimizations such as subscriber admission control, subscriber allocation, broker capacity planning, and broker network planning. There are several factors, which are termed dynamism in this dissertation, that affect the performance of distributed content-based publish/subscribe systems. The proposed analytical framework first addresses each type of dynamism separately in order to avoid the modeling complexity and to study the effect of each type of dynamism individually. The proposed analytical model then relaxes each assumption and combine several types of dynamism altogether under one integrated framework.There are three major types of dynamism considered in the analytical framework : content dynamism, overlay dynamism, and mobility dynamism. Content dynamism means the uncertainty in determining the amount of data from an arbitrary publisher to an arbitrary subscriber due to the publisher-subscriber decoupling nature of the content-based publish/subscribe systems. Overlay dynamism means the uncertainty from publish/subscribe internal broker network, including broker failures and link failures. Finally, mobility dynamism refers to the uncertainty from users' changes of location and content interest. We first propose a probabilistic analytical model for each type of dynamism separately before discussing the framework that integrates all separate analytical models together. We also present validation results for each dynamism-specific analytical model, which prove the accuracy and effectiveness of its corresponding analytical model.This thesis makes contributions in the following areas. First, it proposes a detailed analytical model of content-based publish/subscribe systems from all possible aspects, providing a complete analysis in systematic manner. Second, it incorporates delay and reliability into one single analytical framework, which makes it suitable for delay-sensitive publish/subscribe applications. Third, it discusses and proposes some examples of possible publish/subscribe optimizations on top of such analytical model. Finally, it proves the applicability of the proposed analytical model via simulations with both synthetic and real-world traces.