SMARTNESS was present at the IEEE Conference on Network Function Virtualization and Software Defined Networks (NFV-SDN) 2024 edition in Natal, Brazil. The IEEE NFV-SDN is a forum for the international community to discuss the latest advancements, trends, research breakthroughs and industry applications, and research endeavors in Network Function Virtualization (NFV), Software-Defined Networking (SDN), and related technologies. This event will bring together experts, researchers, and industry leaders worldwide to exchange ideas and explore cutting-edge innovations shaping the future of networks. You can find the program followed during this year’s event here.
SMARTNESS researchers were highly active throughout the congress, presenting two papers, three demos, one doctoral symposium, one tutorial, and one keynote. Following is a review of each work and some of the awards received.
PAPERS:
– RESISTING: A New Fast-Reroute Mechanism with Packet Distribution on P4-Programmable Switches
Authors Daniel De Lima, Francisco Vogt, Alan Teixeira da Silva, Fabricio Rodriguez Cesen, and Christian Esteve Rothenberg (all from UNICAMP) introduce Equal Cost MultiPath (ECMP). This fast re-routing solution optimizes bandwidth efficiency in network infrastructure by using a hash algorithm on the switch’s data plane to distribute packet flows across multiple paths. However, the hash-based approach alone does not address link failures, making deploying a failure recovery solution critical to prevent service disruption and degradation. Fast Reroute (FRR) mechanisms enable rapid recovery when a link fails on the data plane, and integrating FRR with ECMP provides a more resilient solution. This paper introduces RESISTING, a novel ECMP-FRR mechanism for programmable data plane devices. We evaluate our method against the state-of-the-art FRR mechanism on programmable switches by simulating up to three link failures. The results demonstrate that RESISTING achieves equal distribution of packet flows across available links, with no packet losses.
– MATADOR: ML-based Cloud Gaming Traffic Detection entirely in Programmable Hardware
Suneet Kumar Singh (UNICAMP), Christian Esteve Rothenberg (UNICAMP), Alireza Shirmarz (UFSCar), Fábio Luciano Verdi (UFSCar), Israat Haque (Dalhousie University), Gyanesh Patra (Ericsson Research), and Gergely Pongrácz (Ericsson Research) present MATADOR, a machine learning system implemented in P4 programmable hardware to classify cloud gaming (CG) traffic efficiently and accurately. Tested on Tofino switches, MATADOR achieves around 97% accuracy and significantly reduces CG detection time compared to other methods, enhancing the gaming experience by minimizing latency and jitter.
DEMOS:
– Programmable Network Testbed for QoS/QoE Assessment of Holographic Media Delivery
Alan Teixeira da Silva, Rafael P. Silva Clerici, Fabricio E. Rodriguez Cesen, Md Tariqul Islam, and Christian Esteve Rothenberg (all from UNICAMP) present a programmable testbed aimed at assessing the Quality of Service (QoS) and Quality of Experience (QoE) in holographic media delivery. This testbed supports research into how network QoS parameters affect QoE metrics for Point Cloud video on Head Mounted Devices (HMDs), addressing the complex requirements of Holographic Type Communications (HTC) in immersive media beyond 5G.
– PINT-BoX: Path-aware networking IN a Tofino BoX (Best Demo Award)
Everson S. Borges (UFES/IFES), Willen B. Coelho (UFES/IFES), Fabricio R. Cesen (UNICAMP), Francisco G. Vogt (UNICAMP), Christian Rothenberg (UNICAMP), Rodolfo S. Villaça (UFES), Cristina K. Dominicini (IFES), Rafael S. Guimarães (IFES), and Magnos Martinello (UFES) present PINT-BoX, a path-aware networking demonstration using the PolKA source routing approach on a Tofino-based platform. In this collaborative work between SMARTNESS and the FAPESP-funded PORVIR-5G project, the demo explores a path-aware networking approach to improve the performance and resilience of programmable networks. Using the Tofino programmable switch as a high-performance network emulator (check out the github repository of P7), PINT-BoX enables real-time network path management, intelligently responding to changes in traffic and optimizing network resource utilization.
– POSMAC: Powering Up In-Network AR/CG Traffic Classification with Online Learning
Authors Alireza Shirmarz (UFSCar), Fábio Luciano Verdi (UFSCar), Suneet Kumar Singh (UNICAMP), and Christian Esteve Rothenberg (UNICAMP) present POSMAC, a platform that deploys Decision Tree (DT) and Random Forest (RF) models on the NVIDIA DOCA DPU, leveraging an ARM processor for real-time network traffic classification. Designed for Augmented Reality (AR) and Cloud Gaming (CG) traffic, POSMAC enables efficient model evaluation and generalization while optimizing throughput to approach line-rate speeds.
DOCTORAL SYMPOSIUM:
– QoE Evaluation for Emerging Media Applications: Network-Level Analysis and Traffic Modeling (Best PhD Symposium Award)
Authors Md Tariqul Islam (Candidate, UNICAMP) and Christian Esteve Rothenberg (Advisor, UNICAMP) present research on QoE evaluation for Emerging Media Applications (EMA), which include mobile and immersive services in eXtended Reality (XR). The work addresses the challenges of delivering satisfactory QoE over 5G and beyond networks, focusing on network-level measurement, traffic modeling, and optimization to meet high computational and resource demands. This PhD research outlines methodologies and potential research plans for each area, and the work won the Best PhD Symposium Award.
TUTORIALS:
– Recapping your programmable Infrastructures in support of 6G Research: from network emulation to traffic generation
Authors Fabricio Rodriguez (UNICAMP), Francisco Vogt (UNICAMP), Filipo Costa (UNICAMP), Alan Teixeira (UNICAMP), Marcelo Caggiani Luizelli (UNIPAMPA), Christian R. E. Rothenberg (UNICAMP) presented the tutorial Recapping your Programmable Infrastructures in Support of 6G Research. This tutorial demonstrated tools developed by the SMARTNESS team for experimenting with 5G and 6G networks. It introduced P4 and programmable data planes, followed by the P7 tool for emulating 5G/6G networks with varied link characteristics. The team then showcased PIPO-TG for traffic injection, simulating 6G patterns like Time Sensitive Networking (TSN) and Deterministic Networking (DetNet). The integration of these tools enabled a fully emulated environment for network-slicing research.
KEYNOTE:
In his keynote at the MOBISLICE 7 Workshop, Prof. Dr. Christian Esteve Rothenberg (Principal Investigator of the SMARTNESS) presented Software Defined Inception: Turning P4/Tofino Programmable Hardware into an Emulation and Traffic Generation Toolbox for Network Slicing Research. The talk covered the fundamentals of Software-Defined Networking (SDN) and introduced the concept of “Software-defined Inception,” focusing on the use of P4 programmable hardware for network slicing research. He discussed key tools for emulation (P7), traffic generation (PIPO-TG, P4R, and Packet Mancer), and the experimental environments necessary for testing network slicing. The session concluded with practical use cases and further reading on the subject.
