Enhancing perfSONAR Measurement Capabilities using P4 Programmable Data Planes
Ali Mazloum, Jose Gomez, Elie F. Kfoury, Jorge Crichigno · 2023
As a key element in the Science Demilitarized Zone (Science DMZ) framework, perfSONAR is a crucial tool for monitoring and troubleshooting network performance. This paper presents a system that leverages the flexibility and granularity of P4 programmable data planes to enhance perfSONAR measurements. P4 is a programming language used to define how network packets are processed in the data plane. The proposed system employs a P4 programmable data plane that operates passively on real traffic. This approach ensures real-time per-flow monitoring and detailed insights, assisting administrators in understanding network traffic patterns. Moreover, the system detects microbursts and seamlessly integrates with a regular perfSONAR node. Experimental evaluations show that the proposed system provides full visibility of the real traffic, reports undetectable problems by a regular perfSONAR node, and notifies administrators if the endpoints of the connection cause low performance.