Periodic Reporting for period 1 - DAEMON (Network intelligence for aDAptive and sElf-Learning MObile Networks)
Période du rapport: 2021-01-01 au 2021-12-31
DAEMON is a 36-month Research and Innovation Action (RIA) funded by the European Union H2020 framework programme under call H2020-ICT-2018-20. The project studies solutions that will help Beyond-5G (B5G) mobile networks meet high expectations in terms of support for very diverse services, near-zero latency, apparent infinite capacity, and 100% reliability and availability. The main goal of DAEMON is developing and implementing innovative and pragmatic approaches to the design of the Network Intelligence (NI) that will orchestrate physical resources and Virtual Network Functions (VNFs) in next-generation mobile systems. Notably, DAEMON intends to depart from the current hype around Artificial Intelligence (AI) as the silver bullet for any mobile network management task, and realize a systematic integration of varied Machine Learning (ML) models into network architectures that ensures an effective operation of NI in B5G infrastructures.
The project activities are structured around the following set of technical objectives.
Objective 1 – Designing a “NI-native architecture” for B5G systems. DAEMON aims at proposing updates to the network architecture so that it natively supports NI operations. This objective stands on two pillars: (i) enhancing the current architectural vision in B5G standardization efforts so that it implements a well-structured support for coordinated NI across domains, and (ii) fostering a tighter incorporation of NI into the network infrastructure, bringing it closer to the end user and hardware.
Objective 2 – Developing specialized NI-assisted network functionalities for B5G systems. DAEMON designates a concrete list of eight key network functionalities that span a range of operation timescales, network planes and domains. For all these functionalities, the project will devise and implement NI algorithms able of taking advantage of the proposed NI-native architecture, populating the proposed NI-native architecture with actual intelligence.
Objective 3 – Establishing fundamental guidelines for a pragmatic design of NI. DAEMON plans to innovate the way NI is conceived, by leveraging emerging trends in AI and re-thinking how AI is applied to the network infrastructure environment. To this end, DAEMON (i) drafts guidelines on the most appropriate ML tools for specific networking tasks, (ii) designs loss functions for AI model training that are tailored to the network management goal, and (iii) develops AI models that can adaptively trade off accuracy for network-relevant metrics like latency or complexity.
The project activities are structured around the following set of technical objectives.
Objective 1 – Designing a “NI-native architecture” for B5G systems. DAEMON aims at proposing updates to the network architecture so that it natively supports NI operations. This objective stands on two pillars: (i) enhancing the current architectural vision in B5G standardization efforts so that it implements a well-structured support for coordinated NI across domains, and (ii) fostering a tighter incorporation of NI into the network infrastructure, bringing it closer to the end user and hardware.
Objective 2 – Developing specialized NI-assisted network functionalities for B5G systems. DAEMON designates a concrete list of eight key network functionalities that span a range of operation timescales, network planes and domains. For all these functionalities, the project will devise and implement NI algorithms able of taking advantage of the proposed NI-native architecture, populating the proposed NI-native architecture with actual intelligence.
Objective 3 – Establishing fundamental guidelines for a pragmatic design of NI. DAEMON plans to innovate the way NI is conceived, by leveraging emerging trends in AI and re-thinking how AI is applied to the network infrastructure environment. To this end, DAEMON (i) drafts guidelines on the most appropriate ML tools for specific networking tasks, (ii) designs loss functions for AI model training that are tailored to the network management goal, and (iii) develops AI models that can adaptively trade off accuracy for network-relevant metrics like latency or complexity.
To date, DAEMON has performed the following work.
WP2 has focused on four main tasks: (i) drafting functional and performance requirements from the target NI-assisted functionalities; (ii) outlining NI orchestrator and NI interfaces to integrate NI natively in B5G systems; (iii) identifying relevant AI techniques for concrete networking problems, and understanding their limitations; (iv) developing methodologies for novel AI-based approaches that are tailored to real-world NI problems. These tasks produced a list of initial requirements for the DAEMON framework and its NI functionalities, and a literature review of AI for NI that includes initial analyses on customized AI for NI.
In WP3, the work has focused on two main aspects: (i) drafting a functional architecture for real-time NI; (ii) designing specific algorithms that populates the architecture above. These activities have produced a definition of required interfaces for real-time NI, and initial designs of 13 different NI-based algorithms for real-time control of network functions.
In WP4, four parallel and intertwined research threads are pursued: (i) studying energy-aware VNF placements; (ii) designing capacity forecasting models to assisting NI functionalities; (iii) developing a clean-slate automated anomaly detection approach targeting large-scale IoT networks; (iv) developing zero-touch solutions for resource allocation decisions in MANO systems. These activities have resulted in published results related to VNF orchestration for the operation of virtualized base stations, to the deployment of IoT applications in edge infrastructures, to the anticipatory allocation of resources for maximizing end-user Quality of Experience (QoE), and to solutions for Deep-RL-based computation offloading, zero-regret slice reservation and VNF auto-scaling.
In WP5, the following tasks have been carried out: (i) identifying the main simulation/emulation tools that are used in the project in order to validate the NI solutions; (ii) identifying the available sites/testbeds that are available in the project and can be used for the prototyping activities and for providing the project demos; (iii) identifying the available datasets that can be used during the validation process. Synergies with the activities in WP3 and WP4 were established, and a set of over 20 performance evaluation activities are presently ongoing.
WP2 has focused on four main tasks: (i) drafting functional and performance requirements from the target NI-assisted functionalities; (ii) outlining NI orchestrator and NI interfaces to integrate NI natively in B5G systems; (iii) identifying relevant AI techniques for concrete networking problems, and understanding their limitations; (iv) developing methodologies for novel AI-based approaches that are tailored to real-world NI problems. These tasks produced a list of initial requirements for the DAEMON framework and its NI functionalities, and a literature review of AI for NI that includes initial analyses on customized AI for NI.
In WP3, the work has focused on two main aspects: (i) drafting a functional architecture for real-time NI; (ii) designing specific algorithms that populates the architecture above. These activities have produced a definition of required interfaces for real-time NI, and initial designs of 13 different NI-based algorithms for real-time control of network functions.
In WP4, four parallel and intertwined research threads are pursued: (i) studying energy-aware VNF placements; (ii) designing capacity forecasting models to assisting NI functionalities; (iii) developing a clean-slate automated anomaly detection approach targeting large-scale IoT networks; (iv) developing zero-touch solutions for resource allocation decisions in MANO systems. These activities have resulted in published results related to VNF orchestration for the operation of virtualized base stations, to the deployment of IoT applications in edge infrastructures, to the anticipatory allocation of resources for maximizing end-user Quality of Experience (QoE), and to solutions for Deep-RL-based computation offloading, zero-regret slice reservation and VNF auto-scaling.
In WP5, the following tasks have been carried out: (i) identifying the main simulation/emulation tools that are used in the project in order to validate the NI solutions; (ii) identifying the available sites/testbeds that are available in the project and can be used for the prototyping activities and for providing the project demos; (iii) identifying the available datasets that can be used during the validation process. Synergies with the activities in WP3 and WP4 were established, and a set of over 20 performance evaluation activities are presently ongoing.
The current and future activities of DAEMON aim at advancing the state-of-the-art in the design and operation of 8 NI functionalities distributed across different micro-domains of the next-generation mobile architecture, and across controllers, orchestrators and functions that operate at different timescales. Year 1 (Y1) has finished with a number of contributions in this regard from WP3 (see D3.1) and WP4 (see D4.1) published at major conferences and journals.
In addition to solving the challenges inherent to each individual functionality, the project set out to coordinate all this NI within a novel NI Orchestration Layer. A major effort has been devoted from WP2 to this aspect, with a thorough analysis of all NI use cases within scope of the project that produced very valuable insights on requirements, state of the art, and means of verifications of each use case (see D2.1) which shall serve as guidelines for WP3 and WP4. During the second half of Y1, WP2 has striven to plant the foundations for the design of the NI Orchestration Layer.
During Y1 of the project, the partners also worked towards the definition of the tools required to validate the KPIs targeted by DAEMON, including simulators, data-driven emulators, experimental testbeds. This has been a priority in WP5, which kicked off in Q3, and will be reported in the upcoming D5.1. This effort follows a plan of major experimental evaluations, assessing NI for sustainable vRANs, VNF placement and control, anomaly detection and response, edge orchestration, and capacity forecasting and self-learning.
Finally, during Y1, the partners started the work needed to meet the target KPIs of the action in terms of industrial and scientific impact. This is a paramount goal of DAEMON, ensuring that its advancements contribute to relevant industry fora, standards bodies and aid in producing patents. This also helps the long-term impact of the results, and their adoption in industry. As representative figures, during Y1 the project produced (see D6.2 due at M13): 34 scientific publications, including ACM SIGCOMM, IEEE INFOCOM, ACM MobiCom; Eight press releases and magazine articles; Two scientific workshops; One special session of EuCNC 2021; Contributions to one 5G PPP whitepaper; Editing of one international journal special issue; Contributions to two webinars.
All artifacts generated by the project are publicly available via a dedicated community on Zenodo, and linked in the OpenAIRE repository for quick and open access.
In addition to solving the challenges inherent to each individual functionality, the project set out to coordinate all this NI within a novel NI Orchestration Layer. A major effort has been devoted from WP2 to this aspect, with a thorough analysis of all NI use cases within scope of the project that produced very valuable insights on requirements, state of the art, and means of verifications of each use case (see D2.1) which shall serve as guidelines for WP3 and WP4. During the second half of Y1, WP2 has striven to plant the foundations for the design of the NI Orchestration Layer.
During Y1 of the project, the partners also worked towards the definition of the tools required to validate the KPIs targeted by DAEMON, including simulators, data-driven emulators, experimental testbeds. This has been a priority in WP5, which kicked off in Q3, and will be reported in the upcoming D5.1. This effort follows a plan of major experimental evaluations, assessing NI for sustainable vRANs, VNF placement and control, anomaly detection and response, edge orchestration, and capacity forecasting and self-learning.
Finally, during Y1, the partners started the work needed to meet the target KPIs of the action in terms of industrial and scientific impact. This is a paramount goal of DAEMON, ensuring that its advancements contribute to relevant industry fora, standards bodies and aid in producing patents. This also helps the long-term impact of the results, and their adoption in industry. As representative figures, during Y1 the project produced (see D6.2 due at M13): 34 scientific publications, including ACM SIGCOMM, IEEE INFOCOM, ACM MobiCom; Eight press releases and magazine articles; Two scientific workshops; One special session of EuCNC 2021; Contributions to one 5G PPP whitepaper; Editing of one international journal special issue; Contributions to two webinars.
All artifacts generated by the project are publicly available via a dedicated community on Zenodo, and linked in the OpenAIRE repository for quick and open access.