Deploying Machine Learning (ML) models in the user plane enables line-rate in-network inference, reducing latency and enhancing scalability for tasks like traffic monitoring. Yet, integrating ML models into programmable network devices requires meeting stringent constraints in terms of memory resources and computing capabilities. Existing approaches have focused on implementing monolithic ML models within individual programmable network devices, which are limited by hardware capabilities. We propose DUNE, a novel framework enabling, for the first time, distributed user-plane inference across multiple programmable devices. DUNE automates (i) breaking large ML models into smaller sub-models that preserve accuracy while minimizing resource usage, and (ii) designing the sub-models and their sequencing so as to enable an efficient distributed execution of joint packet- and flow-level inference. Implemented in P4 and deployed on a multi-device experimental network, DUNE reduces per-device resource usage and improves inference accuracy by up to 7.5% compared to monolithic designs.

About Beyza Bütün

Beyza Bütün is a Ph.D. student in the Networks Data Science Group at IMDEA Networks Institute in Madrid, Spain. She is also a Ph.D. student in the Department of Telematics Engineering at Universidad Carlos III de Madrid, Spain. She earned both her bachelor’s and master’s degrees in Computer Engineering from Middle East Technical University in Ankara, Turkey, between 2015 and 2022. During her master’s, she worked on the optimal design of wireless data center networks. Beyza’s current research interests are machine learning-based inference, distributed in-band network intelligence, and energy consumption modeling and optimization in the data plane.

This event will be conducted in English