Cost-effective, Energy-efficient, and Scalable Storage Computing for Large-scale AI Applications

Abstract

The growing volume of data produced continuously in the Cloud and at the Edge poses significant challenges for large-scale AI applications to extract and learn useful information from the data in a timely and efficient way. The goal of this article is to explore the use of computational storage to address such challenges by distributed near-data processing. We describe Newport, a high-performance and energy-efficient computational storage developed for realizing the full potential of in-storage processing. To the best of our knowledge, Newport is the first commodity SSD that can be configured to run a server-like operating system, greatly minimizing the effort for creating and maintaining applications running inside the storage. We analyze the benefits of using Newport by running complex AI applications such as image similarity search and object tracking on a large visual dataset. The results demonstrate that data-intensive AI workloads can be efficiently parallelized and offloaded, even to a small set of Newport drives with significant performance gains and energy savings. In addition, we introduce a comprehensive taxonomy of existing computational storage solutions together with a realistic cost analysis for high-volume production, giving a good big picture of the economic feasibility of the computational storage technology.

Past, Present and Future of Computational Storage: A Survey

“…The overview is shown in table 5. Active SSD [1] 09-02-2011 HPC Prototype Active Flash [17,89] 16-04-2012 HPC Simulation Smart SSD [28,38] 06-05-2013 HPC Prototype Intelligent SSD [6,20] 10-06-2013 HPC Simulation Ibex [97] 15-07-2014 Databases Prototype Willow [79] 06-10-2014 Operating Systems (OS) Prototype Biscuit [33] 18-06-2016 Computer Architecture (CA) Prototype Hadoop ISC* [62] 28-07-2016 HPC Prototype YourSQL [37] 15-08-2016 Databases Prototype Caribou [36] 01-08-2017 Databases Prototype Summarizer [41] 14-10-2017 Computer Architecture (CA) Prototype NDP RE2 regex* [13] 11-06-2018 Databases Prototype Registor [64] 26-03-2019 Storage Prototype Cognitive SSD [26] 10-07-2019 Machine Learning Prototype INSIDER [71] 15-07-2019 Storage Prototype Catalina [90] 10-08-2019 HPC Prototype THRIFTY [34] 11-02-2020 Computer Aided Design (CAD) Simulation POLARDB [18] 14-02-2020 Storage Deployment LeapIO [48] 09-03-2020 CA / OS Simulation CSD 2000 [91] 15-09-2020 Storage Prototype NGD newport [27] 12-10-2020 Storage Deployment blockNDP [9] 07-12-2020 Middleware Prototype QEMU CSD* [96] 26-04-2021 Storage Simulation…”

In-storage Processing of I/O Intensive Applications on Computational Storage Drives

“…The processor also provides an environment for running different types of applications. More details of the hardware architecture are available on our previous works [16], [17].…”

A Tapered Floating Point Extension for the Redundant Signed Radix 2 System Using the Canonical Recoding

“…The Booth technique, proposed in 1950 [21], can speed up operations in binary arithmetic by introducing negativeinteger-valued digits. 2 This allows streaks of one's to be eliminated, which reduces the number of partial products generated during multiplication. A natural extension of the Booth technique is the canonical recoding, also known as the nonadjacent form, defined to be a form in which no digit 1 or 1 (negative 1) is adjacent to either a digit 1 or a digit 1.…”

“…In particular, machine learning has become widely adopted as a tool for scientific, industrial, and commercial applications, including autonous systems. This has raised concerns about the energy-and resource-consuming nature of the computationally intensive training of neural networks [1], and the energy and economic cost of data storage, including numerical data, by cloud storage service providers [2], [3]. This has contributed to a recent revival of research into tapered floating point number systems [4]- [6].…”