High Speed Network Processing for Scientific Computing in the Open Cloud Testbed

Abstract

This paper presents an FPGA-based SmartNIC architecture designed to address critical network challenges in data-intensive scientific computing. As scientific workflows increasingly demand high-bandwidth data movement and real-time monitoring across wide-area networks, the Open Cloud Testbed (OCT) provides an ideal platform for investigating in-network processing solutions at line rate. Our approach focuses on implementing network visibility, optimization, and orchestration capabilities through FPGA acceleration. The proposed system leverages P4-programmable data-plane functions to support packet parsing, marking, telemetry extraction, traffic steering, and machine learning inference, all without requiring host CPU intervention or modifications to application servers. As a concrete example, our framework integrates the SciTags standard, which enables associating network traffic with scientific context through packet marking. We implement a high-performance network bridge that counts packets by their SciTag values, as identified in the IPv6 Flow Label. Experimental evaluation on AMD U280 FPGAs in OCT demonstrates the feasibility of this approach, showing imperceptible processing latency and line-rate packet handling, while accurately monitoring scientific data flows.