Evaluation of SmartNIC Devices for Use in Trigger and Data Acquisition Systems

This article presents an evaluation of SmartNIC devices in the context of Trigger and Data AcQuisition (TDAQ) systems. SmartNIC devices represent an emerging technology whose aim is to offload network tasks and infrastructure control plane software from the CPU. Such devices are particularly relevant for TDAQ systems where high rates in the orders of TB/s are produced in large detectors such as the deep underground neutrino experiment (DUNE). In this context, the potential use case of SmartNICs is to perform a quasi-real-time reduction of the incoming data streams by identifying only the interesting signals. The goal is to sustain a number of ~10 Gb data streams aggregated on 100 Gb interfaces and transmit the results to the host machine. An application was developed to provide a testing environment, measuring the achieved throughput in two cases. In the first case, only the total throughput is considered, and the workload is evenly distributed across the available hardware. In the second case, a constraint of processing a number of discrete data streams is added. The application was shown to handle up to ~130 Gbps of incoming data when distributing the workload evenly on the available hardware resources of 8 CPU cores. In this contribution, we show the testing results, optimizations, and hardware tuning of the technology when performing a workload suitable for TDAQ applications.