Resource-Aware Data Analysis

Presented on 30 Aug at DATA 2014

Keynote Lecturer: Katharina Morik

Abstract: Algorithms are designed in consideration of restricted resources. These have been computing time, memory space, and computing nodes. With big data, large computing centers and cloud computing have become popular, but they do not diminish the requirements for space- and time-efficient algorithms. The programming scheme of map and reduce is tailored for processing big data in large computer centers. In contrast, sensor networks, smartphones and other mobile devices cannot rely on a stable access to a computer center, hence stressing the memory constraints and bringing to attention additional resources, namely communication bandwidth and energy. Resource-aware data analysis brings together cyber-physical systems and big data analytics. Learning algorithms are designed to exploit massively parallel hardware, to process data in a stream, and to interact with central capacities in a smart, adaptive way.
This talk presents some results on resource-aware data analysis for smart phones. On the one hand, apps are inspected with respect to their energy consumption. This allows users to adapt their behavior to available resources. However, we want to move beyond that. The operating system, the memory organization, and the wireless communication of devices are to be guided by predictions. Based on predictions, the right time to send and receive data, to prefetch files into cache, and to bundle files for submission can be determined, so to maximize the battery duration of smart phones.