X-TWINUSD: DESIGN AND ARCHITECTURE OF A DISTRIBUTED OPENUSD RUNTIME FOR INTERACTIVE DIGITAL TWIN SIMULATION
- Details
- Hits: 14
Volume 8, Article e2026.03, 2026, Pages 1-16
Elliott Ahn
Gwangju Institute of Science and Technology, Korea, This email address is being protected from spambots. You need JavaScript enabled to view it.
Abstract
Industrial-scale digital twins require interactive visualization and simulation over scene graphs that exceed the memory and metadata capacity of a single node while evolving continuously. OpenUSD provides layered composition semantics, yet naïve stage population and time-varying updates can amplify I/O, saturate metadata services, and inflate memory footprints at scale. This paper presents the design of X-TwinUSD, a distributed runtime architecture that preserves OpenUSD composition semantics while enabling selective payload residency, shard-scoped composition, and incremental delta application for mixed geometry/telemetry workloads. We introduce (i) a Stage-as-Graph representation used to partition a stage into composable shards, (ii) a payload-driven working-set protocol and lifecycle for bounded memory, (iii) an epoch/delta log model for incremental recomposition under bounded delay and out-of-order updates, and (iv) a cost-model interface for scheduling composition and data movement on HPC resources. As a design paper, we focus on the system model, architecture, and correctness considerations, and we additionally report a reproducible microbenchmark demonstrating 10–20× lower time-to-first-interaction, 3–7× lower update latency (p50), and one to three orders-of-magnitude fewer metadata operations under mixed geometry/telemetry workloads.
Keywords:
OpenUSD, Digital twin, HPC systems, Distributed composition, Payload working set, Incremental updates, Microbenchmark evaluation
DOI: doi.org/10.32010/26166127.2026.03
Reference
ADIOS2 (20 December 2025). ADIOS 2 Documentation. https://adios2.readthedocs.io
Ahmad, Y., Kennedy, O., Koch, C., & Nikolic, M. (2012). Dbtoaster: Higher-order delta processing for dynamic, frequently fresh views. arXiv preprint arXiv:1207.0137.
Godoy, W. F., et al. (2020). Adios 2: The adaptable input output system: A framework for high-performance data management. SoftwareX, 12, 100561.
Gupta, A., Mumick, I. S., & Subrahmanian, V. S. (1993). Maintaining views incrementally. ACM SIGMOD Record, 22(2), 157–166.
Naef, M., Lamboray, E., Staadt, O., & Gross, M. (2003, May). The blue-c distributed scene graph. In Proceedings of the Workshop on Virtual Environments 2003 (pp. 125–133).
NVIDIA (20 December 2025). References and Payloads – Learn OpenUSD. https://docs.nvidia.com/learn-openusd/latest/creating-composition-arcs/references-payloads/index.html
NVIDIA (21 December 2025). Platform Overview. Omniverse Developer Overview. https://docs.omniverse.nvidia.com/dev-overview/latest/platform-overview.html
OpenUSD (20 December 2025). Introduction to USD. OpenUSD Documentation. https://openusd.org/docs/
OpenUSD (21 December 2025). Maximizing USD Performance. OpenUSD Documentation. https://openusd.org/docs/Maximizing-USD-Performance.html
OpenUSDa (20 December 2025). USD Terms and Concepts (Glossary). OpenUSD Documentation. https://openusd.org/dev/glossary.html
OpenUSDa (21 December 2025). Pcp: PrimCache Population (Composition). OpenUSD Developer API Reference. https://openusd.org/dev/api/pcp_page_front.html
OpenUSDb (20 December 2025). UsdPayloads Class Reference. OpenUSD API Documentation. https://openusd.org/dev/api/class_usd_payloads.html
OpenUSDb (21 December 2025). Hydra 2.0 Getting Started Guide. OpenUSD Developer API Reference. https://openusd.org/dev/api/_page__hydra__getting__started__guide.html
The HDF Group (20 December 2025). A Brief Introduction to Parallel HDF5. HDF5 Documentation. https://support.hdfgroup.org/documentation/hdf5/latest/_intro_par_h_d_f5.html.
