CERN Accélérateur de science

Published Articles
Title A FPGA-based Network Interface Card with GPUDirect enabling realtime GPU computing in HEP experiments
Author(s) Lonardo, Alessandro (INFN, Rome) ; Ameli, Fabrizio (INFN, Rome) ; Ammendola, Roberto (INFN, Rome2) ; Biagioni, Andrea (INFN, Rome) ; Cotta Ramusino, Angelo (Ferrara U. ; INFN, Ferrara) ; Fiorini, Massimiliano (Ferrara U. ; INFN, Ferrara) ; Frezza, Ottorino (INFN, Rome) ; Lamanna, Gianluca (Frascati ; CERN) ; Lo Cicero, Francesca (INFN, Rome) ; Martinelli, Michele (INFN, Rome) ; Neri, Ilaria (Ferrara U. ; INFN, Ferrara) ; Paolucci, Pier Stanislao (INFN, Rome) ; Pastorelli, Elena (INFN, Rome) ; Pontisso, Luca (INFN, Pisa) ; Rossetti, Davide (Unlisted, US, CA) ; Simeone, Francesco (INFN, Rome) ; Simula, Francesco (INFN, Rome) ; Sozzi, Marco (CERN ; INFN, Pisa) ; Tosoratto, Laura (INFN, Rome) ; Vicini, Piero (INFN, Rome)
Publication 2015
Number of pages 6
DOI 10.3204/DESY-PROC-2014-05/16
Subject category Detectors and Experimental Techniques
Abstract The capability of processing high bandwidth data streams in real-time is a computational requirement common to many High Energy Physics experiments. Keeping the latency of the data transport tasks under control is essential in order to meet this requirement. We present NaNet, a FPGA-based PCIe Network Interface Card design featuring Remote Direct Memory Access towards CPU and GPU memories plus a transport protocol offload module characterized by cycle-accurate upper-bound handling. The combination of these two features allows to relieve almost entirely the OS and the application from data tranfer management, minimizing the unavoidable jitter effects associated to OS process scheduling. The design currently supports one GbE (1000Base-T) and three custom 34 Gbps APElink I/O channels, but four-channels 10GbE (10Base-R) and 2.5 Gbps deterministic latency KM3link versions are being implemented. Two use cases of NaNet will be discussed: the GPU-based low level trigger for the RICH detector in the NA62 experiment and the on- /off-shore data acquisition for the KM3Net-IT underwater neutrino telesco
Copyright/License CC-BY-NC-2.0

Corresponding record in: Inspire


 Notice créée le 2015-08-14, modifiée le 2017-07-31


Fichiers:
Télécharger le document
PDF