blockNDP: Block-Storage Near Data Processing

Antonio Barbalace, Martin Decky, Javier Picorel, Pramod Bhatotia

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract / Description of output

We introduce blockNDP, a set of hardware and software protocol and interface extensions to enable block-oriented near data processing (NDP) within emerging solid-state drives (SSDs) integrating processing units. blockNDP is an end-to-end architecture, including a programming model and framework for host and in-storage compute units. blockNDP is the first proposal to be generic, backward compatible - supports existent standards and file systems, portable, and language agnostic. B@We have implemented and evaluated the effectiveness of blockNDP using two full-system prototypes; one on a new NDP emulation platform - developed atop QEMU, and another on an actual development board - the OpenSSD. Our experimental evaluation demonstrates that blockNDP is not only feasible with a modest effort, < 10K LoC, but it is also able to achieve up to 4 speedups, despite wimpy CPU cores, on a real-world data-driven application such as MySQL.
Original languageEnglish
Title of host publicationProceedings of the 21st International Middleware Conference Industrial Track
Place of PublicationNew York, NY, USA
PublisherACM Association for Computing Machinery
Number of pages8
ISBN (Print)9781450382014
Publication statusPublished - 7 Dec 2020
Event21st International Middleware Conference - Virtual Conference
Duration: 7 Dec 202011 Dec 2020

Publication series

NameMiddleware '20
PublisherAssociation for Computing Machinery


Conference21st International Middleware Conference
Abbreviated titleMiddleware 2020
CityVirtual Conference
Internet address


Dive into the research topics of 'blockNDP: Block-Storage Near Data Processing'. Together they form a unique fingerprint.

Cite this