Open Collections

Context-dependent exogenous coordination for building large scale, dynamic fog computing applications

University of British Columbia

Recently, fog computing has emerged as a new system model for applications that are data-intensive or delay-sensitive. Thanks to the computing infrastructure that is closer to the network edge, communication cost and response time are significantly reduced. While promising, fog computing has its unique challenges, particularly due to its large number of computing elements, their geographic distribution and the dynamic nature of the edge network. First, the large scale complexity of fog computing poses new challenges in the application development process. By analysing various application models, we provide a new taxonomy with important trade-offs that help fog computing developers to navigate their design space in building fog applications. From here, we found that exogenous coordination, where there is a clear separation of concerns between computation and communication activities, is a suitable approach in supporting the complexity of many fog computing scenarios. Second, the geo-distribution of fog devices introduces new applications that depend on such devices' physical context. While we found exogenous coordination to be a suitable approach, it is a software-focused concept that does not cater specifically to those hardware contexts. We propose to extend exogenous coordination with coordination primitives that help to express context-dependent fog application logic. Our proposal includes a clear separation of concerns between contextual and application data, context-dependent constraints for fog application components, and context-bounded communication cardinality among them. Third, the dynamic nature of the edge network requires certain degrees of system monitoring and adaptation, which are resource consuming. To support our proposed coordination model in such a dynamic environment, we develop an incremental coordination technique that based on historical coordination activities to reduce resource consumption. In addition, we show that the coordination activity is not generally affected by the system dynamics with regard to the application's overall performance. This is due to the orthogonality-by-design between coordination and computation aspects of the application. Our platform has been implemented and made publicly available through an open source project, which has been evaluated by both industrial and academic researchers. A large scale, lab-based simulation has also been developed to testify the feasibility of the proposed model.

Text

2019-10-15

Attribution-NonCommercial-NoDerivatives 4.0 International

http://hdl.handle.net/2429/71915

Electrical and Computer Engineering

University of British Columbia

UBCV

http://creativecommons.org/licenses/by-nc-nd/4.0/