The International Conference for High Performance Computing, Networking, Storage, and Analysis
The International Conference for High Performance Computing, Networking, Storage, and Analysis
Authors: Antonino Tumeo (Pacific Northwest National Laboratory (PNNL)), John Feo (Retired), Timothy Mattson (Merly.ai), José Moreira (IBM Thomas J. Watson Research Center), Marco Minutoli (Pacific Northwest National Laboratory (PNNL))
Abstract: Graph analytics is critical to scientific computing, artificial intelligence (AI), and national-scale data analysis. This BoF gathers the community developing high-performance systems for graph processing to discuss current capabilities, emerging challenges, and integration with graph databases, AI workflows, and scientific applications. We will explore both combinatorial and algebraic approaches, including updates from the GraphBLAS community. A key focus is identifying what capabilities—such as open, scalable graph toolchains and support for irregular workloads—require federal investment beyond what commercial vendors provide. The session will guide future research, software development, and funding priorities through expert discussion and broad community input.
Long Description: Graph analytics is increasingly central to scientific computing, artificial intelligence (AI), and world-scale data analysis. Graph-based methods naturally represent relationships, structure, and dependencies in complex systems—from molecular interactions and power grids to knowledge graphs, graph databases, and scientific workflows. These workloads stress high-performance computing (HPC) infrastructure due to their irregular, sparse, and often dynamic behavior. This BoF brings together the community advancing scalable graph toolkits, asynchronous runtimes, and algebraic frameworks to examine current technologies, identify emerging demands, and define a shared roadmap for composable, accelerator-aware graph computing.
Held annually at SC since 2017, this BoF consistently attracts more than 100 engaged attendees from academia, national laboratories, and industry. It provides a forum for critical updates and dialogue across the full spectrum of high-performance graph frameworks—including algebraic approaches like GraphBLAS, as well as combinatorial, task-based, and runtime-driven systems. Topics span data structures, algorithms, APIs, scheduling models, and support for emerging accelerator platforms including GPUs, FPGAs, and custom hardware.
Graph analytics has become essential in many scientific domains: multiscale simulations, bioinformatics, climate science, cybersecurity, and materials discovery. These applications require graph methods to scale across the edge-to-HPC continuum, support streaming and attributed graphs, and integrate with simulation and data platforms. Increasingly, graph databases and semantic models are used in scientific knowledge representation, metadata management, and provenance tracking, often in conjunction with high-performance analytic kernels and AI pipelines. In AI, graph techniques enable graph neural networks (GNNs), drive data preprocessing, and underpin knowledge graphs now being integrated into large language models (LLMs). However, tension exists: LLMs often obscure structured semantics, while graph approaches emphasize explicit relationships and reasoning. This BoF will critically examine these synergies and frictions.
The 2025 session will also build on insights from the Dagstuhl Seminar on Holistic Graph Processing (April 2025), which emphasized the need for better integration of algorithms, system software, and domain abstractions. We will ground the discussion in mature and emerging toolkits, including SuiteSparse:GraphBLAS, CombBLAS, LAGraph, GraphIt, Galois, Gunrock, cuGraph, and SHAD (Scalable High-Performance Algorithms and Data Structures). These frameworks span diverse models—from algebraic kernels to distributed, asynchronous runtimes for large-scale graphs.
A key goal of this BoF is to define a forward-looking research and development agenda that bridges AI, HPC, and scientific workflows. We will highlight areas best suited for federal investment—such as from DOE, NSF, or DARPA—including open toolchains, data-structure composability, graph–database integration, and robust support for heterogeneous, dynamic environments. These are long-term needs unlikely to be fully addressed by commercial actors alone.
Finally, the BoF will catalyze discussion around shared abstractions, unifying layers, and architectural support. Coordination with efforts like SparseBLAS, open graph benchmarks, and HPC–AI co-design programs will help shape a sustainable and impactful future for high-performance graph analytics in both AI and science.
Website: https://hpc.pnl.gov/BOF/