akougkas.io

WisIO: Automated I/O Bottleneck Detection with Multi-Perspective Views for HPC Workflows

Izzet Yildirim, Hariharan Devarajan, Anthony Kougkas, Xian-He Sun, Kathryn Mohror

ICS ’25: 2025 International Conference on SupercomputingACM

Abstract

Modern HPC workloads involve large data transfers that can become bottlenecks. Existing analysis tools identify bottlenecks from per-file performance data but have limitations in parallelizability and rigid heuristic-based rules, necessitating an automated, efficient, and multi-perspective solution. We designed an automated tool, WisIO, that enables parallel and distributed analysis of multi-terabyte-scale workflow performance data. WisIO examines performance data from multiple perspectives, uses metric-driven bottleneck classification, and allows extensible mapping of bottlenecks to root causes. Experimental results demonstrate that WisIO’s multiperspective views substantially improve bottleneck coverage, showing an average increase of up to 805× when compared to analyzing performance data from a single perspective. In our performance evaluation, WisIO’s metric-driven classification processed 340K bottlenecks per second, while its reasoning engine handled around 35K bottlenecks per second. In an analysis of five real-world HPC workloads, WisIO demonstrated up to 11× faster analysis time and identified up to 144× more bottlenecks compared to existing solutions.

Finding

WisIO’s multi-perspective views raise bottleneck coverage by an average of up to 805x, and it analyzed five real-world HPC workloads up to 11x faster while identifying up to 144x more bottlenecks than existing solutions.

System

  1. experimental

    WisIO

    Multi-perspective analysis of HPC I/O traces that connects each bottleneck with its root cause.

Browse all papers on WisIO
  1. 2026
    workshop

    GDSight: Per-Operation Cross-Layer Attribution for GPUDirect Storage

    Izzet Yildirim, Xian-He Sun, Anthony Kougkas. 11th International Parallel Data Systems Workshop (PDSW 2026), held in conjunction with SC26

  2. 2026
    poster

    StageOnce: Speculative Cross-Job Data Staging for LLM Agent Fleets

    Izzet Yildirim, Xian-He Sun, Anthony Kougkas. International Conference for High Performance Computing, Networking, Storage, and Analysis (SC 2026)

  3. 2026
    poster

    ToolFuse: Provenance-Aware In-Flight Coalescing of Redundant Agent Tool Calls

    Izzet Yildirim, Xian-He Sun, Anthony Kougkas. International Conference for High Performance Computing, Networking, Storage, and Analysis (SC 2026)

  4. 2026
    poster

    Extending I/O Observability to Explainability for Targeted Optimization

    Neeraj Rajesh, S. Neuwirth, Jay Lofstead, Anthony Kougkas, Xian-He Sun, Nikoli Dryden et al. International Conference for High Performance Computing, Networking, Storage, and Analysis (SC 2026)

Citation

@inproceedings{yildirim_wisio_2025,
  author = {Yildirim, Izzet and Devarajan, Hariharan and Kougkas, Anthony and Sun, Xian-He and Mohror, Kathryn},
  booktitle = {ICS '25: 2025 International Conference on Supercomputing},
  title = {WisIO: Automated I/O Bottleneck Detection with Multi-Perspective Views for HPC Workflows},
  year = {2025},
  month = jun,
  publisher = {ACM},
  pages = {749--763},
  doi = {10.1145/3721145.3725742},
  url = {https://doi.org/10.1145/3721145.3725742},
}

Related notes

  1. essay

    One destination, N tiers

    Every tier supplies the application at once, and the application sees one destination. The thesis from a November 2026 talk to my group at the Gnosis Research Center, the six lines of I/O work behind it since 2016, and my bet that they are the foundation agents need for context.

    9 min
  2. personal

    2025 reflections

    A personal record of 2025: a promotion, a keynote, two dissertations, a summer of teaching, a loss, and the argument I kept making all year about what a researcher is for now.

    7 min