Energy-Optimized Elastic Application Distribution for Automotive Systems in Hybrid Cloud Architectures

Abstract

The increase of resource-intensive applications in modern vehicles used for streaming, gaming or autonomous driving results in rising energy-consumption of its advanced computing and connectivity hardware. Especially in electric vehicles, this leads to much higher hardware costs and a decreased vehicle range. Modern premium cars use distributed heterogeneous hardware and mostly communicate via APIs to large cloud backends. Current approaches to reduce on-board energy consumption offload applications partly and make use of limited network connectivity assumptions to their backends. In this paper, we propose a hybrid electric and electronic architecture that manages vehicle hardware by using cloud computing frameworks. Our hybrid cloud architecture is a connection of the local vehicle cloud and a large data centre community cloud. We propose an online optimization algorithm that shifts applications from on-board ECUs to data centre servers and vice-versa. The optimization algorithm minimizes the local energy-consumption while satisfying predicatively dynamic constraints like data rate limitations, application policies and resource limitations. Our approach outperforms the non-predictive approach in average by 16%, in the best case by 21% and in the worst case both behave equally well.