Authors: Dong Li (University of California, Merced), Da Li (Yotta Labs), Deborah Bard (Lawrence Berkeley National Laboratory (LBNL)), Ian Foster (University of Chicago), Nicholas Malaya (Advanced Micro Devices, Inc. (AMD)), Paolo Faraboschi (Arm Limited)
Abstract: This BoF session will examine the potential of decentralized data center architectures to advance high performance computing (HPC) and artificial intelligence (AI). Discussions will focus on the technical, operational, and policy-driven dimensions of distributed systems, including edge computing, federated learning, and energy-efficient design strategies. The session aims to engage researchers, system architects, and data center professionals in exploring scalable and resilient alternatives to traditional centralized infrastructures. Participants will have an opportunity to exchange insights, share implementation experiences, and foster collaborations that inform the future of computing infrastructure.
Long Description: As HPC and AI workloads continue to grow in scale and complexity, traditional data center architectures are facing critical challenges in sustainability, scalability, and resilience. This BoF will convene a diverse set of stakeholders to explore the concept and implementation of decentralized data centers—a distributed paradigm that leverages edge computing, modular microdata centers, federated resource orchestration, and decentralized governance models.
*Goals*
- Facilitate a dynamic discussion on the architectural shifts needed to support decentralized computing for HPC and AI workloads.
- Exchange experiences and best practices from research labs, academic institutions, industry practitioners, and cloud service providers exploring decentralized models.
- Identify common challenges—such as data sovereignty, security, heterogeneity, and resource scheduling—and brainstorm collaborative solutions.
- Build momentum toward standardization and shared frameworks that can enable interoperability and performance benchmarking across decentralized infrastructures.
*Topics to Discuss*
(1) Architectural models of decentralized data centers for HPC
- Comparative analysis of centralized vs. decentralized data center architectures for HPC/AI
- Deployment models: coordination of multiple data centers, edge compute clusters, and federated supercomputing networks
- Hardware considerations: modular designs, composable infrastructure, high-speed interconnects across distributed nodes
(2) Intelligent Resource Management
- AI-driven scheduling and resource orchestration across geographically dispersed environments
- Decentralized workload balancing using real-time telemetry and predictive modeling
- Handling data locality, replication, and efficient in-transit compute operations
(3) Security Access & Governance
- Identity federation, zero-trust security models, and multi-factor authentication across trust domains
- Secure data movement and encrypted workload execution between institutions
- Governance structures for inter-organizational collaboration (e.g., consortia, co-ownership models)
(4) Interoperability & Standardization
- Role of open-source frameworks in managing decentralized compute (e.g., Kubernetes, Slurm with federation support)
- APIs and interfaces for cross-system communication and monitoring
- Benchmarking standards and reproducibility in distributed environments
(5) Sustainability and Energy Efficiency
- Strategies for lowering the carbon footprint of decentralized infrastructures
- Renewable energy integration: onsite solar, energy-aware workload placement
- Cooling innovations for smaller, regional HPC facilities
(6) Real-Time & Geographically Distributed Use Cases
- Scientific instrumentation data processing at the edge (e.g., radio telescopes, seismology)
- Climate and disaster modeling with globally distributed compute nodes
- AI model training across diverse data sources and regional regulations
(7) Collaboration & Adoption
- Models for funding distributed infrastructure (e.g., shared grants, public-private partnerships)
- Institutional case studies of decentralized deployments
- Challenges in onboarding institutions, aligning technical roadmaps, and sustaining long-term operations
*Relevance to the HPC Community*
This BoF is directly aligned with the SC conference’s mission to advance innovation in high performance computing, networking, storage, and analysis. As data-intensive AI and HPC workloads push the limits of centralized systems, decentralized data centers offer a compelling alternative—particularly in addressing resilience, geographic accessibility, and democratization of compute resources.
Attendees will gain insights into emerging models that could redefine how the HPC community conceptualizes infrastructure ownership, collaboration, and scalability. The BoF also seeks to highlight open-source efforts and cross-institutional initiatives that embody the community spirit of HPC research.
Whether your institution is grappling with resource constraints, environmental sustainability goals, or the need to support real-time, globally distributed computation—this BoF offers a stage to share lessons, raise critical questions, and shape future directions.