A Novel Framework for the Seamless Integration of FPGA Accelerators with Big Data Analytics Frameworks in Heterogeneous Data Centers

Abstract

To face the increased network traffic in the cloud, data center operators have started adopting an heterogeneous approach in their infrastructures. Heterogeneous infrastructures, e.g. based on FPGAs, can provide higher performance and better energy-efficiency compared to the contemporary processors. However, FPGAs lack of an easy-to-use framework for the efficient deployment from high-level programming frameworks. In this paper, we present a novel framework that allows the seamless integration of FPGAs from high-level programming languages, like Java and Scala. The proposed approach provides all the required APIs for the utilization of FPGAs from these languages. The proposed scheme has been mapped on Amazon AWS f1 infrastructure and a performance evaluation is presented for two widely used machine learning algorithms.