Exploiting the Omnisc'IO prediction approach for Exascale systems

Research topic and goals

Many techniques used in data management such as scheduling of I/O requests, prefetching, and caching, can benefit from knowing ahead of time when files will be accessed, and how. Omnisc’IO proposes an approach based on formal grammars to predict such future accesses in HPC simulations. The present research topic aims to improve Omnisc’IO and leverage its predictions in schedulers, burst buffers, and caching systems of future machines to prevent cross-application interference and to improve I/O performance.

Results for 2015/2016

The Omnisc’IO approach was presented at SC 14 (Dorier et al. 2014). An extension of the approach was published in the TPDS journal (Dorier et al. 2015).

Visits and meetings

None.

Impact and publications

  1. Dorier, Matthieu, Shadi Ibrahim, Gabriel Antoniu, and Rob Ross. 2015. “Using Formal Grammars To Predict I/O Behaviors in HPC: the Omnisc’IO Approach.” IEEE Transactions On Parallel and Distributed Systems. Institute of Electrical and Electronics Engineers. doi:10.1109/TPDS.2015.2485980.
    @article{DorierTPDS2015,
      title = {{Using Formal Grammars to Predict I/O Behaviors in HPC: the Omnisc'IO Approach}},
      author = {Dorier, Matthieu and Ibrahim, Shadi and Antoniu, Gabriel and Ross, Rob},
      url = {https://hal.inria.fr/hal-01238103},
      journal = {{IEEE Transactions on Parallel and Distributed Systems}},
      publisher = {{Institute of Electrical and Electronics Engineers}},
      year = {2015},
      doi = {10.1109/TPDS.2015.2485980},
      keywords = {Omnisc'IO ; Grammar ; Prediction ; I/O ; Storage ; HPC},
      pdf = {https://hal.inria.fr/hal-01238103/file/paper.pdf},
      hal_id = {hal-01238103},
      hal_version = {v1}
    }
    
  2. ———. 2014. “Omnisc’IO: A Grammar-Based Approach to Spatial and Temporal I/O Patterns Prediction.” In Proceedings Of the International Conference for High Performance Computing, Networking, Storage and Analysis, 623–34. SC ’14. Piscataway, NJ, USA: IEEE Press. doi:10.1109/SC.2014.56.
    @inproceedings{DorierEtAl2014a,
      acmid = {2683662},
      address = {Piscataway, NJ, USA},
      author = {Dorier, Matthieu and Ibrahim, Shadi and Antoniu, Gabriel and Ross, Rob},
      booktitle = {Proceedings of the International Conference for High Performance Computing,
          Networking, Storage and Analysis},
      doi = {10.1109/SC.2014.56},
      isbn = {978-1-4799-5500-8},
      keywords = {HPC, I/O, Omnisc'IO, exascale, grammar, prediction, storage},
      location = {New Orleans, Louisana},
      numpages = {12},
      pages = {623--634},
      publisher = {IEEE Press},
      series = {SC '14},
      title = {Omnisc'IO: A Grammar-based Approach to Spatial and Temporal I/O Patterns Prediction},
      year = {2014}
    }
    

Future plans

The project is currently suspended due to a lack of human resources.

References