The main goal of the DAEMON project is to develop and implement innovative and pragmatic approaches to network intelligence (NI) design. This project aims to enable a high-performance, sustainable, and extremely reliable zero-touch network system. By departing from the current hype surrounding Artificial Intelligence (AI) as the silver bullet for mobile network management tasks, the DAEMON project seeks to find a more sensible solution based on an in-depth understanding of the PHY pipeline and the requirements of each step.
Departing from AI Hype in Mobile Network Management
While AI has gained significant attention in automating networking scenarios, the DAEMON project argues that it is not a one-size-fits-all solution. By critically evaluating the limitations of AI in mobile network management tasks, the project emphasizes the need for a more comprehensive approach that takes into account the intricacies of the PHY pipeline and the unique requirements of each step in the process.
Risks of Running DUs in Shared Edge Clouds
Within the context of the DAEMON project, the risks of running Distributed Units (DUs) in shared edge clouds are demonstrated. This situation often leads to resource competition, compromising the performance and reliability of the network. Recognizing this challenge, the project highlights the necessity for a better solution that addresses the drawbacks of shared edge cloud deployments.
Nuberu: Overcoming Legacy DU Design Limitations
Presented at the ACM MobiCom 2021 conference, Nuberu is a breakthrough solution that overcomes the limitations of legacy DU designs. This innovative approach, developed within the DAEMON project, offers improved performance, sustainability, and reliability in DU deployments. By leveraging advanced techniques and strategies, Nuberu provides a path towards enhanced network functionality and efficiency.
The role of Machine Learning (ML) models in networking automation has become increasingly important. These models are utilized in various networking scenarios, such as traffic classification, quality of service (QoS) prediction, and routing optimization. The DAEMON project acknowledges the importance of ML models and aims to integrate them effectively into the network architecture to enhance automation processes.
Exploiting Off-the-Shelf Programmable Data Planes
To achieve low-latency and high-throughput inference in networks, the DAEMON project exploits off-the-shelf programmable data planes such as Intel Tofino ASICs. These programmable data planes, combined with domain-specific languages like P4, offer a favorable environment for implementing innovative network infrastructure designs. Their utilization enables efficient network performance and enhances overall system reliability.
Challenges in Embedding ML Models into User Plane Hardware
Embedding ML models into user plane hardware, particularly programmable switches, presents various challenges. The DAEMON project acknowledges the constraints imposed by programmable switches in terms of memory, support for mathematical operations, and the number of allowed per-packet operations. Overcoming these challenges is crucial to effectively implement ML models within production-grade hardware.
Enabling RF Model Embedding in Programmable Switches
To address the challenges associated with incorporating ML models into programmable switches, the DAEMON project introduces Flowrest. This practical framework allows for the execution of RF models at the flow level in real-world programmable switches. Flowrest facilitates the embedding of large RF models into production-grade hardware, thereby enhancing the network’s intelligence and performance.
Addressing Challenges and Test Results
By designing the Flowrest framework and addressing the aforementioned challenges, the DAEMON project demonstrates remarkable improvements in accuracy. Through tests involving tasks of unprecedented complexity, the project showcases how their model can improve accuracy by up to 39% compared to previous approaches to implementing RF models in real-world equipment. These results underscore the effectiveness and practicality of the DAEMON project’s approach to advancing NI design.
The DAEMON project is at the forefront of developing and implementing innovative approaches to network intelligence design. By departing from the AI hype and focusing on practical solutions, the project strives to achieve high-performance, sustainable, and extremely reliable zero-touch network systems. Through groundbreaking advancements like Nuberu and Flowrest, the project tackles the limitations of legacy designs, enables efficient ML model embedding, and significantly improves network accuracy. The DAEMON project holds great promise in shaping the future of network automation and advancing the field of network intelligence.