Institut national de recherche scientifique français Univerité Pierre et Marie Curie Université Paris Diderot - Paris 7

Real-time data acquisition and processing on GPU with applications in radioastronomy and radar systems

jeudi 17 février 2022, par Julien Plante (LESIA)

Jeudi 24 février 2022 à 16h00 , Lieu : En visioconférence

Data throughput in modern telescopes has been quickly increasing over the last decade, with bandwidths as high as tens of terabits per second expected with the Square Kilometer Array. A similar trend can be observed with radar systems.

Moreover, a frequent requirement in these systems is to process the stream of raw data from sensors, sent using transport protocols based on e.g. UDP, in real-time. To be able to process information at high throughput and sometimes low latency, accelerators like GPUs or FPGAs are popular solutions, but are still complex to use.

During my PhD, I am investigating the use of commercial off-the-shelf network interfaces to perform the data ingestion in a more standard and easy way, with applications on telescopes and on radar systems. I will present my progress with the DPDK library and the GPUDirect technology provided by Nvidia, and how this work is being integrated with the COSMIC framework developed in HRAA. This could provide easier data acquisition for any application in the future.