ENIAD: A Reconfigurable Near-data Processing Architecture for Web-Scale AI-enriched Big Data Service

Abstract

To meet the surging demands required by AI-enriched Big Data services, cloud vendors are turning toward domain specific accelerators for improved efficiency, scalability and performance. ENIAD, the first end-to-end infrastructure for AI-enriched Big Data serving in real time, accelerates both deep neural network inferencing and billion-scale indexing at the data-center scale. Exploiting near- data computation, reconfigurable computing and rapid/agile hardware deployment flow, ENIAD serves state-of-the-art, online built indexing service with high efficiency at low batch sizes. A high-performance, index (data)-adaptable FPGA soft processor is at the heart of the system and able to serve 10x larger index size with 14x lower latency compared to state-of-the-art CPU and GPU architectures.