Spark acceleration on FPGAs: A use case on machine learning in Pynq

Abstract

Spark is one of the most widely used frameworks for data analytics. Spark allows fast development for several applications like machine learning, graph computations, etc. In this paper, we present Spynq: A framework for the efficient deployment of data analytics on embedded systems that are based on the heterogeneous MPSoC FPGA called Pynq. The mapping of Spark on Pynq allows that fast deployment of embedded and cyber-physical systems that are used in edge and fog computing. The proposed platform is evaluated in a typical machine learning application based on logistic regression. The performance evaluation shows that the heterogeneous FPGA-based MPSoC can achieve up to 11× speedup compared to the execution time in the ARM cores and can reduce significantly the development time of embedded and cyber-physical systems on Spark applications.