D3.2 Report on infrastructure resources, access and use policies, v1

Deliverable D3.2 offers the initial overview of available computing infrastructure resources that support AI application development and execution within SAIFA. It also establishes a common framework for accessing and using these resources. During the first implementation phase, data was gathered on national HPC and AI resources supporting SAIFA activities, including their technical specifications, capacity, access procedures, and resources allocated to the project. At the same time, the team collected and reviewed the national resource providers' existing access and usage policies. To facilitate early-stage development, SAIFA researchers now have direct access to national computing resources managed by the Office for Information Technologies and eGovernment (ITE), the Institute of Physics Belgrade (IPB), and the Innovation Center at the School of Electrical Engineering, University of Belgrade (ICEF). Currently, each resource provider grants access directly, with the project providing support and coordination. Users must adhere to the policies and conditions governing each resource. This setup is temporary until access is unified via the SAIFA AI Gateway. Based on the analysis of the policies applied by the national resource providers, SAIFA has developed a common set of use policies covering the principal requirements related to resource use, user responsibilities, security, data management, privacy and compliance. The objective is to establish a consistent policy framework applicable across resources accessed through SAIFA, while allowing individual providers to impose additional resource-specific conditions where necessary. This common framework will be further evaluated and aligned with the policies of the collaborating European AI Factories (Pharos and IT4LIA) as their resources become available to SAIFA users.In parallel, we have developed a common procedure for requesting and allocating computing resources. Its implementation will be gradual: the current direct access arrangements support immediate development needs, while the procedure will ultimately be fully incorporated into the SAIFA AI Gateway, which will provide a common entry point for requesting and managing access to the available infrastructure. Resource providers will retain responsibility for operating their infrastructure and applying resource-specific technical, security, and allocation requirements. The overall resource model differentiates between infrastructure dedicated to development activities and that supporting production-scale operations. National resources will mainly aid in the development, integration, testing, and preparation of SAIFA applications and workflows. For resource-intensive production runs, SAIFA will utilize resources supplied by the collaborating PHAROS and IT4LIA AI Factories, as they become available. Additionally, access to other EuroHPC systems will supplement these resources and will be managed separately, with usage levels dictated by application needs and the availability and capacity of the PHAROS and IT4LIA resources.This deliverable thus sets the foundational connection between existing infrastructure, operational access setups, and a unified policy framework. As the SAIFA AI Gateway becomes active and European AI Factory resources are incorporated into the service offerings, the access procedures and shared use policies will be further improved and harmonized to meet the needs of the relevant infrastructure providers.