Identify HPC software components that benefit from NIC offload

Determine which components of the high-performance computing (HPC) software stack measurably benefit from offloading communication and data-processing tasks to smart network interface cards (sNICs), in practical end-to-end deployments.

Background

The paper introduces FPsPIN, an FPGA-based, open-hardware implementation of the sPIN abstract machine model that enables packet-level processing and offload on smart NICs. While prior work has shown benefits in simulations, the authors emphasize the need for full-system evaluations to understand practical gains and limitations across real workloads.

In this context, the authors explicitly pose a broad open question aimed at the HPC community: determining which parts of the software stack (e.g., MPI collectives, datatype processing, or other middleware components) actually benefit from NIC-based offload when evaluated end-to-end with realistic systems and constraints.