28th Conference on Computing in High Energy and Nuclear Physics (CHEP 2026)

A Path Toward Scalable FPGA Offloading in Federated Research Infrastructures

Not scheduled

18m

Chulalongkorn University

Poster Presentation Track 7 - Computing infrastructure and sustainability Poster

Speakers

Daniele Spiga Diego Ciangottini (INFN, Perugia (IT)) Francesco Brivio (Universita & INFN, Milano-Bicocca (IT)) Giulio Bianchini (Universita e INFN, Perugia (IT)) Massimo Sgaravatto (Universita e INFN, Padova (IT)) Mirko Mariotti (Universita e INFN, Perugia (IT)) Paolo Dini Simone Gennai (Universita & INFN, Milano-Bicocca (IT))

Description

In the last years, INFN has consolidated and expanded its distributed computing infrastructure toward heterogeneous hardware systems, also thanks to the ICSC and TeraBIT projects, funded in the context of the Italian National Recovery and Resilience Plan. Among the most innovative components of the national federation are the specialized hardware clusters known as HPC Bubbles, in particular those equipped with FPGAs and GPUs.
This contribution describes the work done to integrate FPGA resources hosted in the Milano Bicocca Bubble and resource-rich nodes hosted in Padova. Relying on the interLink technology, Kubernetes-based deployments (hosted on the INFN Cloud at CNAF) are enabled both to transparently execute computationally intensive firmware synthesis and build workflows exploiting large CPU and memory capacity clusters, and to access FPGA devices for the execution of externally synthesized firmware.
The approach demonstrates the feasibility of decoupling firmware build and execution phases across heterogeneous infrastructures and enables systematic experimentation with heterogeneous FPGA offloading across multiple Kubernetes domains. The current status and early results obtained through the integration with the AI_INFN platform will be shown.