Periodic Reporting for period 1 - DESIRE6G (Deep Programmability and Secure Distributed Intelligence for Real-Time End-to-End 6G Networks)
Reporting period: 2023-01-01 to 2024-06-30
Current and envisaged use cases, extending beyond 2030, require a more flexible and dynamic network architecture. 6G systems are expected to further broaden the range of vertical sectors supported, with reduced complexity while pushing performance limits even further. Simplicity should be a primary goal for 6G, achieved as a natural outcome of utilizing cutting-edge technologies such as the ability to program the user plane, cloud-native features, seamless acceleration of network functions (NFs) or application workloads, and data-driven network automation. DESIRE6G proposes a novel architecture leveraging these technologies to support diverse use cases, focusing on ultra-low latency and mission-critical control loops.
The primary objective of DESIRE6G is to design and prototype a 6G network centered on data-driven autonomic networking and deep programmability, enabling rapid service generation, automated optimizations, and easy-to-use APIs towards the applications, etc. A key proposition is the use of an end-to-end programmable data plane supporting multitenancy, providing flexibility in workload offloading and customization of network behavior, considering performance and power efficiency. To this end, DESIRE6G introduces an infrastructure management layer that separates business logic from the infrastructure layer, simplifying the use of hybrid hardware systems and cloud-native resource management.
DESIRE6G is targeting autonomic networking through a architecture that employs an intent-base service management and orchestration layer, introducing Multi-Agent Systems (MAS) for distributed intelligent control, bringing network intelligence closer to the user plane for near-real-time decision-making. This results in an AI-powered multi-level service optimization framework that considers various inputs, KPIs, and policies to optimize services and infrastructure. The project also introduces a pervasive monitoring system, utilizing network telemetry for accurate end-to-end information collection. DESIRE6G employs Distributed Ledger Technology (DLT) as a zero-trust mechanism throughout its architecture, enabling dynamic service federation across multiple administrative domains and enhancing the security of the MAS-based approach.
The primary objective of DESIRE6G is to design and prototype a 6G network centered on data-driven autonomic networking and deep programmability, enabling rapid service generation, automated optimizations, and easy-to-use APIs towards the applications, etc. A key proposition is the use of an end-to-end programmable data plane supporting multitenancy, providing flexibility in workload offloading and customization of network behavior, considering performance and power efficiency. To this end, DESIRE6G introduces an infrastructure management layer that separates business logic from the infrastructure layer, simplifying the use of hybrid hardware systems and cloud-native resource management.
DESIRE6G is targeting autonomic networking through a architecture that employs an intent-base service management and orchestration layer, introducing Multi-Agent Systems (MAS) for distributed intelligent control, bringing network intelligence closer to the user plane for near-real-time decision-making. This results in an AI-powered multi-level service optimization framework that considers various inputs, KPIs, and policies to optimize services and infrastructure. The project also introduces a pervasive monitoring system, utilizing network telemetry for accurate end-to-end information collection. DESIRE6G employs Distributed Ledger Technology (DLT) as a zero-trust mechanism throughout its architecture, enabling dynamic service federation across multiple administrative domains and enhancing the security of the MAS-based approach.
During the first 18 months, the project completed the use case definition, requirement analysis and provided the reference DESIRE6G system architecture; a zero-touch orchestration management and control platform, with native integration of AI over a performant, measurable and programable data plane (PDP). With regards to the latter, the project has made significant progress in developing the PDP components. These include the implementation of key IML components; programmable traffic management approaches for deep slicing; programmable and accelerated network functions, focusing on RAN functions and AI workloads; in-network computation as a service. Simultaneously, the project progressed in advancing intelligent orchestration, management and control within the DESIRE6G platform. Efforts focused on defining and implementing the functional components of the Service Management and Orchestration (SMO) framework including the development of the ML Function Orchestrator (MLFO) enabling secure AI/ML pipelines for near-real-time control of 6G Network Services. The design and implementation of the AI/ML agents have been integral to these efforts, ensuring secure interactions and autonomous operations for each service. Additionally, work on the SMO framework included analysing and defining the approach to Intent-Based Networking and the development of an Optimization engine to further support service deployment and assurance. Significant work has also been done on enabling secure, dynamic service federation across multiple domains. Furthermore, DESIRE6G designed and is currently developing a pervasive monitoring architecture, incorporating in band telemetry solutions and (in network) data aggregation, alongside RAN telemetry integration, to support service optimization and assurance. DESIRE6G established the infrastructure and processes necessary for the integration, validation, and demonstration of DESIRE6G's innovations. The key achievements include the setup of the distributed testbed infrastructure, which is composed of two main integration testbeds and several federated testbeds used for component testing and preliminary validation activities. Preliminary PoCs demonstrations were successfully conducted. DESIRE6G has actively contributed to standardization efforts (IETF, 3GPP, O-RAN) and open-source activities, with the consortium delivering several technical contributions
The project:
-Specified the novel DESIRE6G system architecture.
-Developed a modular micro-service architecture for the SMO and introduced a sandboxing mechanism for workloads running on compute resources enabling secure multi-tenancy.
-Introduced a distributed approach for autonomous near-real-time QoS assurance using DRL and MASs to create a distributed, collaborative network control plane.
-Secured the deployed ML pipelines using DLT for key exchange and AI agent attestation.
-Introduced novel AI/ML solutions employing centralized knowledge for non-RT decision-making, distributed knowledge for Near-RT decision-making and edge intelligence.
-Introduced fast, secure, and dynamic service federation across different domains using DLT(blockchain and smart contracts)
-Designed an Infrastructure Management Layer for managing resources, including hardware-accelerated data planes (ASICs, DPUs, smartNICs, FPGAs), that supports multi-tenancy for P4 programmable data planes.
-Designed/implemented programmable traffic management solutions.
-Identified 11 network functions for hardware acceleration, starting the development on 4, and introduced load optimization methods that employ hybrid CPU and hardware-accelerated NF instances.
-Started the integration of the SOL framework with the vAccel framework to provide a pure cloud-native approach for ML model inference.
-Adopted in-band telemetry on the top of the existing postcard telemetry solutions for monitoring.
-Developed a low-overhead software monitoring tool with minimal performance penalty.
-Built a distributed D6G testbed infrastructure and performed preliminary PoCs and demos.
-Released open-source data sets for AI training and inference and demonstrated early Y2 demos at conferences.
The results have been disseminated via 25 scientific papers, 30 talks, 5 demonstrators, 8 standards and 5 open-source contributions, and 5 patent applications.
-Specified the novel DESIRE6G system architecture.
-Developed a modular micro-service architecture for the SMO and introduced a sandboxing mechanism for workloads running on compute resources enabling secure multi-tenancy.
-Introduced a distributed approach for autonomous near-real-time QoS assurance using DRL and MASs to create a distributed, collaborative network control plane.
-Secured the deployed ML pipelines using DLT for key exchange and AI agent attestation.
-Introduced novel AI/ML solutions employing centralized knowledge for non-RT decision-making, distributed knowledge for Near-RT decision-making and edge intelligence.
-Introduced fast, secure, and dynamic service federation across different domains using DLT(blockchain and smart contracts)
-Designed an Infrastructure Management Layer for managing resources, including hardware-accelerated data planes (ASICs, DPUs, smartNICs, FPGAs), that supports multi-tenancy for P4 programmable data planes.
-Designed/implemented programmable traffic management solutions.
-Identified 11 network functions for hardware acceleration, starting the development on 4, and introduced load optimization methods that employ hybrid CPU and hardware-accelerated NF instances.
-Started the integration of the SOL framework with the vAccel framework to provide a pure cloud-native approach for ML model inference.
-Adopted in-band telemetry on the top of the existing postcard telemetry solutions for monitoring.
-Developed a low-overhead software monitoring tool with minimal performance penalty.
-Built a distributed D6G testbed infrastructure and performed preliminary PoCs and demos.
-Released open-source data sets for AI training and inference and demonstrated early Y2 demos at conferences.
The results have been disseminated via 25 scientific papers, 30 talks, 5 demonstrators, 8 standards and 5 open-source contributions, and 5 patent applications.