Power-Efficient Data Processing with Software-Defined Computational Storage

webinar

Author(s)/Presenter(s):

Yang Seok Ki

Library Content Type

Presentation

Library Release Date

Focus Areas

Abstract

CPU performance improvements based on Dennard scaling and Moore's Law have already reached their limits, and domain-specific computing has been considered as an alternative to overcome the limitations of traditional CPU-centric computing models. Domain-specific computing, seen in early graphics cards and network cards, has expanded into the field of accelerators such as GPGPUs, TPUs, and FPGAs as machine learning and blockchain technologies become more common. In addition, hyperscalers, where power efficiency is particularly important, use ASICs or FPGAs to offload and accelerate OS, security, and data processing tasks. Meanwhile, technologies such as cloud, machine learning, big data, and the edge are generating data explosively, and the recent emergence of high-performance devices such as over-hundred-gigabit networks, NVMe SSDs and SCMs has made CPU-centric computing more the bottleneck. Processing large amounts of data in a power-efficient manner requires re-examining the existing model of moving data from storage to the CPU, which consumes a lot of power and limits performance due to bandwidth limitations. Eventually, we expect that each device will extend the functions it performs into the realm of computing per its needs, and each device will participate in heterogeneous computing coordinated by the CPU. Samsung believes that near-data processing, or in-storage computing is another important piece of the puzzle. In this keynote, we look back at the system architecture that has changed to handle a variety of data and discuss the changes we expect from system architecture in the future. And we'll talk about what Samsung can contribute to these changes, including the evolution of computational storage, form factors, features, roles, benefits, and components. We'll also look at the ecosystem elements this computational storage needs to settle into, and talk about areas in which various industry players need to work together.

  • Datacenter architecture
  • Domain specific computing
  • Computational storage architecture