Data Management Challenges of Exascale Scientific Simulations: A Case Study with the Gyrokinetic Toroidal Code and ADIOS
The push towards exascale computing and the recent introduction of multi-petascale supercomputers have enabled science applications to run complex simulations. However, the gap between compute and I/O has grown wider, even as applications seek to generate and persist increasing amounts of data. Optimizing I/O is challenging and remains a bottleneck at scale. In this paper, we present initial I/O performance results of running Gyrokinetic Toroidal Code (GTC) on Summit, a 200 Petaflop system at Oak Ridge National Laboratory. To manage the complex data in GTC, we use ADIOS, an I/O and data management middleware that provides a rich set of APIs to manage and interact with scientific data. We discuss optimizations performed to obtain improvements in I/O performance and identify a set of challenges that will drive the design and development of next generation data management libraries.
- Research Organization:
- Oak Ridge National Lab. (ORNL), Oak Ridge, TN (United States)
- Sponsoring Organization:
- USDOE
- DOE Contract Number:
- AC05-00OR22725
- OSTI ID:
- 1558473
- Resource Relation:
- Conference: 10th International Conference on Computational Methods (ICCM 2019) - Singapore, , Singapore - 7/9/2019 8:00:00 AM-7/13/2019 8:00:00 AM
- Country of Publication:
- United States
- Language:
- English
Similar Records
On the energy footprint of I/O management in Exascale HPC systems
...And Eat it Too: High Read Performance in Write-Optimized HPC I/O Middleware File Formats