The DiPET project is an ambitious, 3-year collaboration between five renowned institutions that investigates the dynamic and transparent mapping of stream processing applications on fog and edge computing environments. The end goal is to improve operational aspects as indicated by resource utilization and energy consumption, and to enhance user experience.
Work Package 1 (WP1), Lead: UVT
WP1 aims to develop streaming algorithms with tunable precision for use in data intensive workloads in edge/fog systems. We will develop novel transprecise operators as well as providing methodologies for parameter tuning machine learning algorithms.
Work Package 2 (WP2), Lead: FORTH
WP2 aims to define and implement a programming model on top of Apache Flink APIs to allow application developers to express transprecise streaming computations (business logic) separately from scheduling constraints and placement policies.
Work Package 3 (WP3), Lead: IRISA
WP3 aims to dynamically map transprecise stream processing applications to the set of available fog computing resources and precision requirements that can deliver the requested quality of service while minimizing the number of exploited resources. It then continuously adjusts this mapping to account for variations in the application demands and the availability of fog computing resources.
Work Package 4 (WP4), Lead: UPC
WP4 aims at evaluating the DiPET components in terms of technology, business exploitation and societal impact. Technology will be evaluated through simulations, and real field experiments in the two use case scenarios, while business exploitation and societal impact will be investigated through analysis and modeling.