Hardware Acceleration Landscape for Distributed Real-time Analytics: Virtues and Limitations
Mohammadreza Najafi, Kaiwen Zhang, Hans-Arno Jacobsen and Mohammad Sadoghi
Technische Universitat Munchen, Technische Universitat Munchen, University of Toronto, Purdue University

Arguably, we are now witnessing a new technological revolution with the potential that ranges from transforming our day-to-day life experiences (e.g., personalized medicine and education) to transforming every single industry (e.g., data-driven healthcare, commerce, agriculture, and mining). At the core of this revolution lies data. This transformation is facilitated by sensing, gathering, and connecting all physical entities to construct a rich and dynamic computational model of reality in real-time. Every procedure and every decision needed in the physical world will soon be optimized in real-time by ingesting and analyzing massive volume of present and past data at an unprecedented velocity. To cope with such extreme scale, we argue the need to revisit the hardware and software co-design in light of two key technological advancements. First is the virtualization of computation and storage over highly distributed data centers spanning across continents. Second is the emergence of a variety of specialized hardware accelerators that complement the traditional general-purpose processors. We argue there is an imminent need to exploit and unify these two trends in order to unleash and harness the power of data in real-time. In this paper, we focus on presenting a formulation and characterization of hardware acceleration landscape geared towards real-time analytics in the cloud environment. Our goal is to assist both researchers and practitioners navigating the newly revived field of software and hardware co-design for building next generation distributed systems. We further present a case study to explore software and hardware interplay for designing distributed realtime stream processing.