This position paper presents a novel heterogeneous CPU-GPU multi-level cloud acceleration focusing on applications running on embedded systems found on low-power devices. A runtime system performs energy and performance estimations in order to automatically select local CPU-based and GPU-based tasks that should be seamlessly executed on more powerful remote devices or cloud infrastructures. Moreover, it proposes, for the first time, a secure unified model where almost any device or infrastructure can operate as an accelerated entity and/or as an accelerator serving other less powerful devices in a secure way.
|Number of pages||4|
|Journal||Procedia Computer Science|
|Publication status||Published - 17 Oct 2016|
|Event||Cloud Futures: From Distributed to Complete Computing, CF2016, 18-20 October 2016, Madrid, Spain - Madrid, Spain|
Duration: 18 Oct 2016 → 20 Feb 2017