GRC at SC26 (opens in a new tab)
GRC researchers will present eleven accepted works at SC26 from November 16 to 19, 2026, and Anthony Kougkas will co-lead a Birds of a Feather session. All times are CST. On Monday, November 16, at PDSW'26, the 11th International Parallel Data Systems Workshop (room W175bc), Fernandes et al. will present GPUH5, a GPU-first I/O library for scientific data (10:50 to 11:10am); Yildirim et al. will present GDSight, per-operation cross-layer attribution for GPUDirect Storage (11:30 to 11:50am); Xu et al. will give a five-minute talk on trigger-gated multimodal LLM agents for in situ simulation monitoring (12:15 to 12:20pm); and Teves et al. will give a five-minute talk on type-safe, dynamically defined metadata for scientific applications (4:45 to 4:50pm). The same morning, at ProTools'26, the 8th Workshop on Programming and Performance Visualization Tools (room W179b), Devarajan et al. will present Orion (10:30 to 10:54am) and DataCrumbs (10:54 to 11:18am), both with GRC coauthors. On Tuesday, November 17, four research posters will be on display in North Pre-Function from 10:00am to 5:00pm: ToolFuse and StageOnce by Yildirim et al., I/O observability extended to explainability by Rajesh et al., and GNNs Beyond HBM by Pawar et al. On Wednesday, November 18, Xu et al. will present the paper "Trigger, Render, Reason: Multimodal LLM Agents for Science-Aware Scientific Simulation Monitoring" in the session Papers in Conversation: Scientific AI at Extreme Scale (4:30 to 5:00pm, room W181abc). On Thursday, November 19, the Birds of a Feather session "Trustworthy Data at Scale: HDF5 in the Age of AI-Driven Science" will meet from 12:15 to 1:15pm in room W185d, led by Michael Breitenfeld of The HDF Group with Elena Pourmal, Suren Byna, Quincey Koziol, and Anthony Kougkas.
- PaperGPUH5: A GPU-first I/O Library for Scientific Data
- PaperGDSight: Per-Operation Cross-Layer Attribution for GPUDirect Storage
- PaperTrigger-Gated Multimodal LLM Agents for In-situ Scientific Simulation Monitoring
- PaperType-Safe, Dynamically Defined Metadata Support for Scientific Applications
- PaperOrion: Observability framework for hybrid workflow-centric I/O monitoring in HPC systems
- PaperDataCrumbs: Efficient Cross-Layer Data Capture for Explainable I/O Performance in HPC Storage Stacks
- PaperToolFuse: Provenance-Aware In-Flight Coalescing of Redundant Agent Tool Calls
- PaperStageOnce: Speculative Cross-Job Data Staging for LLM Agent Fleets
- PaperExtending I/O Observability to Explainability for Targeted Optimization
- PaperGNNs Beyond HBM: Compressed, Bit-Exact Feature Streaming for Out-of-Core Training
- PaperTrigger, Render, Reason: Multimodal LLM Agents for Science-Aware Scientific Simulation Monitoring