D4.2 - Distributed Network Design

The rapid advancements in Artificial Intelligence (AI) and Machine Learning (ML) are driving innovation across various domains, from autonomous systems to personalized medicine. However, achieving the full potential of these technologies requires robust research infrastructures capable of supporting distributed, scalable, and secure AI development. dAIEDGE aims to establish a cuttingedge virtual research lab that addresses these challenges by enabling seamless collaboration, resource sharing, and AI model development across geographically distributed environments. In this context, Task T4.2 focuses on designing a distributed network for the dAIEDGE virtual research lab, leveraging blockchain technology as a core enabler for decentralization, security, and trust. This task aims to establish the architecture and technical foundation needed to interconnect diverse research infrastructures, facilitate remote access to resources, and create a secure and scalable environment for AI research and development. Objectives of Task T4.2 The primary objective of Task T4.2 is to design a blockchain-based distributed network for the virtual research lab. This design process encompasses the identification and specification of functional and non-functional requirements, the development of a formal architecture model, and the definition of an Open API that will enable seamless interconnection and aggregation of remote research assets. Specific objectives include: • Designing a Blockchain-Enabled Research Network: Leverage blockchain to provide decentralized coordination, secure interactions, and transparent access control across the distributed infrastructure. • Defining Functional and Non-Functional Requirements: Establish clear requirements that will guide the design of the architecture and Open API, ensuring scalability, interoperability, security, and performance. • Developing a Formal Architecture Model: Propose an architecture that integrates the blockchain network with orchestration solutions (e.g., Kubernetes) to optimize AI deployment, provisioning, and resource management. • Collaborating Across Stakeholders: Engage with key partners to align the design with existing platforms such as the Bonseyes Marketplace, the AIoD platform roadmap, and other research infrastructures. Expanding the Design via Use Case Implementation in T4.4 The design and architecture model developed in Task T4.2 will be further expanded and refined through the implementation of specific use cases in Task T4.4. This iterative approach will involve close collaboration with project partners, who will implement and validate the architecture in realworld scenarios. Through this process, the Formal Architecture Model and Functional and NonFunctional Requirements will be continually developed and enhanced, informed by practical feedback and insights gathered during these implementations. An overview of the three use cases are below: Use Case 1 : Benchmarks: Enhancing HW/SW Benchmarking for AI at the Edge This use case leverages and extends the benchmarking suite developed in the Bonseyes project, focusing on evaluating AI applications on edge and deep edge devices. The benchmarking process includes assessing performance, power efficiency, and scalability across a variety of hardware/software configurations, ranging from low-power microcontrollers to more capable edge processors. The objectives are to: o Validate AI Workflows on Diverse Edge Devices: Measure how well AI models perform when deployed on various edge platforms, considering both computeconstrained devices and more capable edge infrastructure. o Optimize Hardware/Software Co-Design: Generate insights that guide the cooptimization of AI models, software frameworks, and edge hardware configurations, ensuring efficient deployment. o Drive Standardization: Provide benchmarks that contribute to industry-standard metrics for evaluating AI at the edge, facilitating wider adoption of best practices. This use case is crucial for refining the virtual research lab’s capability to support edge AI deployment, ensuring that the architecture is robust, scalable, and adaptable to different hardware environments. Use Case 2 : Protection of User Assets: Secure AI Asset Management In this use case, the focus is on testing a framework that integrates encryption mechanisms for safeguarding sensitive AI Assets, such as models, datasets, and proprietary algorithms designed to establish trust and integrity within the virtual research lab. The framework includes: • Keyless Code Signing: Implementing secure code-signing mechanisms that allow AI assets, including models and software components, to be verified without relying on traditional key-based systems. This approach reduces complexity and enhances scalability in distributed environments. • Attestations and Provenance: Introducing mechanisms for securely recording and verifying the origin and integrity of AI assets. This includes tracking the history of data, models, and algorithms as they are developed, shared, and deployed across the lab, ensuring traceability and accountability. • Secure Data Handling: Integrating security protocols that protect sensitive AI assets and data throughout their lifecycle, ensuring compliance with privacy regulations and minimizing the risk of unauthorized access. This use case aligns directly with the security requirements guiding the architecture, ensuring that the virtual lab supports trusted, secure collaboration among stakeholders. Use Case 3 : Experimentation, Industrialization, and Interoperability This use case focuses on bridging the gap between research prototypes and real-world edge applications by enabling the transition of AI assets from experimental stages to industrial deployment. The key activities include: • AI Asset Integration and Deployment: Providing a pipeline for taking AI models and solutions developed within the virtual lab and deploying them on edge devices, with a particular focus on platforms based on RISC-V architecture. • Interoperability and Platform Integration: Ensuring that AI Assets are interoperable across different edge computing environments and platforms. This involves close collaboration with relevant AI platforms to achieve seamless integration. • Industrialization of AI Workflows: Supporting the move from experimental setups to scalable, production-ready systems, including the automation of deployment, monitoring, and updates. Adopting a use case-driven approach to design and validation offers several key advantages: 1. Real-World Validation: By grounding the design in concrete use cases, the architecture model and requirements are more likely to address real-world needs, ensuring the virtual lab is relevant, practical, and aligned with industry demands. 2. Iterative Refinement: Use case implementations allow for an iterative design process where feedback loops from real deployments can be integrated back into the architecture. This leads to continuous improvement and ensures that the final solution is both robust and flexible. 3. Comprehensive Requirement Gathering: Use cases help in capturing both functional and non-functional requirements that may not be fully evident in a purely theoretical design. For example, performance bottlenecks, security issues, and deployment challenges can be identified and resolved through practical testing. 4. Cross-Domain Applicability: The diverse nature of the use cases (from AI benchmarking to secure code signing and encryption) ensures that the virtual research lab’s design is versatile and adaptable across multiple domains and application scenarios. 5. Stakeholder Engagement: Involving project partners in the implementation and testing phases encourages active engagement and collaboration, leading to a more aligned and consensus-driven outcome. This also helps in leveraging partner expertise in areas like orchestration, security, and edge computing. Through the implementation of these use cases in Task T4.4, the design of the dAIEDGE virtual research lab will be refined to target a TRL level of 3-4. Scope and Structure of the Deliverable This deliverable presents the results of Task T4.2, detailing the design and specification of the virtual research lab. The report is structured as follows: • Section 1: Requirement Analysis: Provides a detailed analysis of the functional and nonfunctional requirements, including security, scalability, and interoperability considerations. • Section 2: Architecture Design: Describes the proposed architecture model for the blockchain-enabled distributed network, highlighting the integration of key components and services. • Section 3: Open API Specification: Outlines the design and functionalities of the Open API, detailing the endpoints, access controls, and integration with edge components. • Section 4: Use Cases: Provides an overview of the specific use cases that guided the design and validation of the architecture. • Section 5: Conclusions and Next Steps: Summarizes the key outcomes and provides an outlook on further developments, including alignment with Task 4.4. The design and specifications presented in this report aim to lay a strong foundation for the development and deployment of the dAIEDGE virtual research lab, ultimately advancing the stateof-the-art in distributed AI research infrastructures. Stakeholder Contributions: • BCA (Task Leader): Leading the requirements and architecture for the VLab, leveraging expertise from the Bonseyes Marketplace and aligning with the AIoD (AI on Demand) platform roadmap. • BTH: Contributing orchestration concepts based on Kubernetes for optimizing ML deployment. • CETIC: Identifying integration requirements with edge-focused virtual labs and cybersecurity-focused software factories. • KUL: Focusing on security requirements and employing security- and privacy-by-design principles. • HES-SO: Expertise on heterogeneous edge systems and dedicated HPC infrastructure. • SED: