OctoRay: Framework for Scalable FPGA Cluster Acceleration of Python Big Data Applications
Zaid Al-Ars, Jakoba Petri-Koenig, Joost Hoozemans, Luc Dierick, H. Peter Hofstee · 2023
In this paper, we introduce OctoRay, a Python framework that allows users to easily combine several FPGA and Python libraries to run their big data analytics pipelines in parallel on FPGA-enabled clusters. We show that OctoRay provides users with multiple levels of freedom, regarding the type of FPGA, the choice of the application and the number of available FPGAs per node, while illustrating the simplicity of the usage of the framework. Additionally, we evaluate the performance of the framework on two use cases that represent typical stages of many big data analytics pipelines. We show that the FPGA implementation outperforms the software baseline by 2x to 12x as the number of FPGAs scales from 1 to 6. This demonstrates that the speedup scales linearly with the number of FPGAs, indicating the speedup potential OctoRay offers and the low overhead of the framework. OctoRay is open source and publicly available at https://github.com/abs-tudelft/octoray