Data Centers

Input/Output APIs and Data Organization for High Performance Scientific Computing

Download Now Free registration required

Executive Summary

Scientific data formats like HDF-5 and NetCDF have given rise to rich tool chains for use by science end users. Performance issues with their direct use by petascale (and beyond) applications, however, demonstrate the need for; improvements in scalability with respect to the degree of attainable parallelism, a specific technique used in this paper being delayed data consistency, the ability to perform rapid data characterization to improve scientific productivity, and resilience to isolated failures through improved data organization.

  • Format: PDF
  • Size: 123.77 KB