RESTful Hardware Microservices Using Reconfigurable Networked Accelerators in Cloud and Edge Datacenters

Abstract

We propose enabling cloud datacenters with Reconfigurable Networked Accelerators RNAs. RNAs are FPGA and memory compute nodes connected to the main network of the datacenter. To enable seamless integration of RNAs, we propose RESTful hardware microservices in cloud datacenters. We show how a front-end model view controller (MVC) web application can issue a call to remote RNA-accelerated RESTful microservices to decrease the latency of a single client query and increase the throughput of clients served. As a use case, we investigate just in time classification of client uploaded media (e.g., images, videos, etc.) against adult or hateful content. The system architecture is implemented using Spring MVC (Spring Boot) and AlexNet convolutional neural network CNN for image classification. Observed results show up to more than 10x improvements in throughput and energy efficiency depending on the target RNA (FPGA) device and the level of optimization of the employed hardware classifier.