Streaming Message Interface: High-Performance Distributed Memory Programming on Reconfigurable Hardware
Identifikátory výsledku
Kód výsledku v IS VaVaI
<a href="https://www.isvavai.cz/riv?ss=detail&h=RIV%2F61989100%3A27740%2F19%3A10242820" target="_blank" >RIV/61989100:27740/19:10242820 - isvavai.cz</a>
Výsledek na webu
<a href="https://dl.acm.org/doi/pdf/10.1145/3295500.3356201" target="_blank" >https://dl.acm.org/doi/pdf/10.1145/3295500.3356201</a>
DOI - Digital Object Identifier
<a href="http://dx.doi.org/10.1145/3295500.3356201" target="_blank" >10.1145/3295500.3356201</a>
Alternativní jazyky
Jazyk výsledku
angličtina
Název v původním jazyce
Streaming Message Interface: High-Performance Distributed Memory Programming on Reconfigurable Hardware
Popis výsledku v původním jazyce
Distributed memory programming is the established paradigm used for high-performance computing (HPC) systems, requiring explicit communication between nodes and devices. When FPGAs are deployed in a distributed setting, communication is typically handled either by going through the host machine, thereby sacrificing performance; or by streaming between devices across fixed point-to-point connections, sacrificing flexibility. We present Streaming Message Interface (SMI), a communication model and API that unifies explicit message passing with a hardware-oriented programming model, facilitating minimal-overhead, flexible, and productive inter-FPGA communication. SMI utilizes a dedicated FPGA interconnect with runtime-configurable routing, and supports arbitrary network topologies. Messages are streamed over the network during computation, allowing communication to be easily integrated into pipelined designs. We provide a high-level synthesis implementation of SMI, and use it to implement a set of distributed memory applications. Using SMI, programmers can implement distributed, scalable HPC programs, without deviating from best practices for hardware design.
Název v anglickém jazyce
Streaming Message Interface: High-Performance Distributed Memory Programming on Reconfigurable Hardware
Popis výsledku anglicky
Distributed memory programming is the established paradigm used for high-performance computing (HPC) systems, requiring explicit communication between nodes and devices. When FPGAs are deployed in a distributed setting, communication is typically handled either by going through the host machine, thereby sacrificing performance; or by streaming between devices across fixed point-to-point connections, sacrificing flexibility. We present Streaming Message Interface (SMI), a communication model and API that unifies explicit message passing with a hardware-oriented programming model, facilitating minimal-overhead, flexible, and productive inter-FPGA communication. SMI utilizes a dedicated FPGA interconnect with runtime-configurable routing, and supports arbitrary network topologies. Messages are streamed over the network during computation, allowing communication to be easily integrated into pipelined designs. We provide a high-level synthesis implementation of SMI, and use it to implement a set of distributed memory applications. Using SMI, programmers can implement distributed, scalable HPC programs, without deviating from best practices for hardware design.
Klasifikace
Druh
D - Stať ve sborníku
CEP obor
—
OECD FORD obor
10201 - Computer sciences, information science, bioinformathics (hardware development to be 2.2, social aspect to be 5.8)
Návaznosti výsledku
Projekt
<a href="/cs/project/EF16_027%2F0008463" target="_blank" >EF16_027/0008463: Věda bez hranic</a><br>
Návaznosti
P - Projekt vyzkumu a vyvoje financovany z verejnych zdroju (s odkazem do CEP)
Ostatní
Rok uplatnění
2019
Kód důvěrnosti údajů
S - Úplné a pravdivé údaje o projektu nepodléhají ochraně podle zvláštních právních předpisů
Údaje specifické pro druh výsledku
Název statě ve sborníku
International Conference for High Performance Computing, Networking, Storage and Analysis, SC
ISBN
978-1-4503-6229-0
ISSN
2167-4329
e-ISSN
—
Počet stran výsledku
33
Strana od-do
33
Název nakladatele
IEEE
Místo vydání
Piscataway
Místo konání akce
Denver
Datum konání akce
17. 11. 2019
Typ akce podle státní příslušnosti
WRD - Celosvětová akce
Kód UT WoS článku
—