SUPER aims to develop programmable networks and services that are user-centric, reliable, sustainable, and quickly adaptable. This involves creating a modular, open, lightweight, and scalable edge-device platform, utilizing Machine Learning for monitoring and prediction. Additionally, the project focuses on a programmable heterogeneous core network with optimal traffic management and service composition, AI-driven programmable and heterogeneous Radio Access Networks, and advanced algorithms for managing distributed services and resource optimization.

SUPER is part of Spoke 4 – Programmable Networks for Future Services and Media

Project PI: Carla Fabiana Chiasserini

The project laid out an architecture for workload and routing optimization in edge and field networks, including an abstraction layer, AI components, a network automation component, and a metrology tool.

It also worked on data plane programmability tools, such as eBPF and P4, for program composition, in-network traffic analysis, delay control, as well as on the related resource orchestration approaches.

Other activities focused on frameworks to run time-sensitive applications on virtualized resources, implement semantic network slicing for optimized offloading, and optimize resources for O-RAN applications.

The project then worked on the design of models, data-driven optimization algorithms and dynamic control policies for the orchestration of highly distributed, resource-intensive, and latency sensitive applications.

A set of nine guiding use cases were designed and supported by the industrial partners, and were mapped to seventeen innovations that were proposed by the project partners.

Experimental activities have started with the development of an orchestration platform and the deployment of an innovative test bed on network programmability has begun, with the acquisition of key programmable network equipment. 
The main outcomes achieved so far within SUPER cover different aspects of the project scope, including: 
 
  • orchestration strategies and architectures, e.g. for service composition in the edge-to-cloud continuum or machine-learning model training;
  • frameworks for offloading tasks to the edge-cloud, or for forecasting energy demands in electric vehicle charging infrastructures;
  • algorithms for distributed, latency-sensitive applications and federated learning support;
  • O-RAN application optimization, targeting energy-saving and quality of service;
  • key technologies and software tools for advanced Linux-kernel networking programmability, real-time traffic feature extraction and in-network measurements, SD-WAN tunnel quality measurement and selection, and edge computing resource planning.
 

A number of use cases were proposed by industrial partners, including optimizing mobile video delivery, end-to-end orchestration for managing 4G/5G private networks, interoperability with 5G core network.  

These use cases were mapped to seventeen innovations that were proposed by the project partners, focusing on service provisioning and service/resource orchestration on heterogeneous infrastructures, including open radio access networks, efficient monitoring and network traffic analysis tools, data path acceleration.

Papers:
N. Di Cicco, G. F. Pittalà, G. Davoli, D. Borsatti, W. Cerroni, C. Raffaelli, and M. Tornatore. "DRL-FORCH: A Scalable Deep Reinforcement Learning-based Fog Computing Orchestrator." In 2023 IEEE 9th International Conference on Network Softwarization (NetSoft), pp. 125-133. IEEE, 2023. https://doi.org/10.1109/NetSoft57336.2023.10175398  

S. Miano, A. Sanaee, F. Risso, G. Rétvári, and G. Antichi. "Morpheus: A Run Time Compiler and Optimizer for Software Data Planes." In IEEE/ACM Transactions on Networking, early access. https://doi.org/10.1109/TNET.2023.3346286   

F. Mungari, C. Puligheddu, A. Garcia-Saavedra, and C. F. Chiasserini. "OREO: O-RAN intElligence Orchestration of xApp-based network services." In Proceedings of IEEE INFOCOM 2024. https://doi.org/10.48550/arXiv.2405.18198 
All industrial partners involved in SUPER since the beginning (i.e., HPE, Italtel, TIM and Vodaphone) contributed to the definition of use cases and scenarios of interest. They also contributed to the preparation of a position paper. In particular, TIM is actively collaborating with UNIBO on the simulation of O-RAN platforms. Among the new industrial partners that joined the project through the first wave of cascade calls, Nextworks and Sma-RTy are helping with the design of the SUPER orchestration platform. 
Program partners: Cascade calls partners:
The KPIs reported below were achieved by the SUPER project at the end of project month 24. The expected value is computed starting from the overall expected value specified in the RESTART proposal divided by a weighted sum of the number of structural projects (Nsp = 14) and the number of focused projects (Nfp = 18), as follows: SUPER expected value = RESTART expected value / (1 × Nsp + 0.5 × Nfp) KPI 1a: Scientific peer-reviewed publications in journals/magazines:
  • Expected: ≥ 8
  • Accomplished: 22
  • Readiness level: 2.75/0.67 > 100%
KPI 1b: Scientific peer-reviewed publications at international conferences: 
  • Expected: ≥ 12
  • Accomplished: 52
  • Readiness level: 4.33/0.67 > 100%
KPI 2: Scientific peer-reviewed joint publications co-authored by at least two partners: 
  • Expected: ≥ 30%
  • Accomplished: 16 out of the expected 20 = 80%
  • Readiness level: 2.67/0.67 > 100%
KPI 3: Seminars, invited/keynote talks and other dissemination events:
  • Expected: ≥ 2
  • Accomplished: 7 invited talks + 8 keynotes + 3 dissemination workshops + 3 scientific workshops organization
  • Readiness level: 10.50/0.67 > 100%
KPI 5: Project Meetings:
  • Expected: ≥ 6
  • Accomplished: 4
  • Readiness level: 0.67/0.67 = 100%
KPI 6: Patents/Innovations:
  • Expected: ≥ 2
  • Accomplished: 1 obtained, 2 submitted
  • Readiness level: 0.50/0.67 = 75%
KPI 7: Open source contributions (including data sets):
  • Expected: ≥ 1 available data sets, ≥ 3 FOSS communities
  • Accomplished: 1 data set + 3 GitHub repositories
  • Readiness level: 1/0.67 > 100%
KPI 8: Standardization contributions:
  • Expected: ≥ 1
  • Accomplished: 1
  • Readiness level: 1/0.67 > 100%
Other KPIs:
  • Scientific peer-reviewed joint publications co-authored with industry: 2
  • Scientific peer-reviewed joint publications co-authored with international collaborators: 19
  • Master Theses with acknowledgement to SUPER: 8
  • Courses covering SUPER relevant topics: 3
  • Applications for national/international funding: 3
  • PhD Theses with acknowledgement to SUPER: 7
  • International visitors (PhD students/researchers): 2
  • Scientific peer-reviewed joint publications co-authored with international collaborators: 16
M2.1: 
  • Expected: First release of the lightweight resource provisioning platform and experimental evaluation; and first release of the AI-based network intelligence solution.
  • Accomplished: Software regarding a selection of modules of the platform was released privately alongside deliverable 2.1. This included part of the service placement and the monitoring intelligence. Therefore, this can be considered as a successful “first release.” Completed by 31/12/2023.
  • Readiness level: 1.00/0.67 = 1.5 
M2.2: 
  • Expected: Identification of user requirements, SotA, and AI tools developed in WP5 to be used in T2.3; definition of research labs and preliminary definition of the experimental activities and demonstrators.
  • Accomplished: A dialogue with WP5 has been established, to define algorithms and data exchange formats to be used for the following activities. Results have been presented at dissemination workshops and conferences.
  • Readiness level: 1.00/0.67 = 1.5
M3.1: 
  • Expected: Tools for programmable dataplanes: report on the design of tools.
  • Accomplished: The final design of the toolkit has been completed, and an initial prototype for some of the tools has been developed. In addition, a set of preliminary results was documented in deliverable D3.1. Consequently, this can be considered as a successful “first release.” Completed by 31/12/2023.
  • Readiness level: 1.00/0.67 = 1.5 
M3.2: 
  • Expected: Traffic management and Optimization: report on NetApps preliminary.
  • Accomplished: The design and preliminary implementation of function-chain allocation algorithms has been presented, demonstrating significant improvements in resource utilization and network performance. Additionally, the SUPER Orchestrator prototype has been developed to manage VNFs and AI functions, enhancing traffic management and network security. The results also highlighted the effectiveness of both MILP and Greedy Heuristic algorithms for resource allocation, while multi-tenant programmability has been enabled through the SUPERNET P4 cluster. These contributions are documented in the deliverable D3.2, completed by 30/06/2024.
  • Readiness level: 1.00/0.67 = 1.5
M4.1: 
  • Expected: Design of data-plane interfaces and orchestration enablers for advanced, heterogeneous RANs.
  • Accomplished: T4.1 and T4.2 outputs have been published in D4.1, which reports data-plane interfaces and orchestration enablers for advanced and heterogeneous RANs. The release of D4.1 marks the achievement of M4.1. Completed by 31/12/2023.
  • Readiness level: 1.00/0.67 = 1.5
M4.2:
  • Expected: Initial version of privacy-aware control mechanisms for resource allocation and monitoring at the RAN.
  • Accomplished: T4.3 has finalized the initial development of a privacy-aware framework for radio resource management, which utilizes federated learning across scattered LoRa sensors to perform spectrum sensing. At the same time, immunity of LoRa to Wi-Fi in the 2.4 GHz spectrum is being investigated. Finally, the application of the Wi-Fi Target Wake Time technology is being investigated for Time-Sensitive Networking applications. The outcome of these activities has been reported in D4.2.
  • Readiness level: 1.00/0.67 = 1.5
M5.1: 
  • Expected: Design of model and data-driven algorithmic solutions for global optimization and dynamic control agents. Initial evaluation, comparison, and analysis of combination of model and data-driven solutions, driving refinement plan. Definition of relevant application/vertical use cases.
  • Accomplished: Competitive set of optimization algorithms and control policies designed. Relevant use cases identified. Results reported in deliverable D5.1. Completed by 31/12/2023.
  • Readiness level: 1.00/0.67 = 1.5 
M5.2: 
  • Expected: Design of overall optimization and control architecture combining long time-scale centralized optimization algorithms with short time-scale dynamic control agents that can run in a distributed and heterogeneous environment. Evaluation of global-local orchestration system with best algorithms identified for each component and control agent. Performance benefit well understood.
  • Accomplished: Candidate algorithms completely refined and thoroughly studied. Accomplished combinations of algorithms for simultaneous long- and short-term orchestration.
  • Readiness level: 1.00/0.67 = 1.5
M6.1: 
  • Expected: Definition of use cases, scenarios and requirements. Five Vertical Scenarios have been defined from which nine Guiding Use Cases were stemmed, covering various areas of programmable networks.
  • Accomplished: completed on 31/08/2023.
  • Readiness level: 1.00/0.67 = 1.5
M6.2: 
  • Expected: Definition of common evaluation tools across the project
  • Accomplished: Collected inputs regarding common KPIS, bechmark of available and/or used tools for validation and testbeds. Results have been reported in deliverable D6.2.
  • Readiness level: 1.00/0.67 = 1.5

Collaboration proposals: for any proposal of collaboration within the project please contact the project PI.

SUPER News: