GISolve - Spatial Interpolation on PRAGMA

Members: Shaowen Wang, Ransom Briggs, Eric Shook
This spatial interpolation application is developed by Shaowen Wang and Ransom Briggs to solve a k-nearest neighbor search problem for inverse distance weighted spatial interpolation. Wikipedia has more information about different types of interpolation and their possible uses.

System Requirements

Network

  • No network requirements are made on the Worker Node, the application runs without contacting any other machine

Hardware and Software Requirements

Hardware
  • x86 compatible CPU
  • 60 MB disk per job
  • 512 MB memory
Software
  • Almost any Linux distribution

Virtual Organization support

Disk Structure

Application Directory ($APP)
  • Meant for reusable application files
  • Mounted read-write on the Compute Element
  • Mounted at the same path on Worker Nodes
Data Directory ($DATA)
  • Meant for temporary data storage until staged out
  • Mounted read-write on the Compute Element and Worker Nodes
  • Mounted at the same path on Worker Nodes
Application life cycle
  • Run fork job to create directories in $DATA for staging in and out
  • Stage application tarballs into $DATA
  • Run fork job to install application from tarball
  • Now the site is ready to run jobs
Job life cycle
  • Stage data into $DATA
  • Run batch job with executable name and arguments
  • Stage data out of $DATA
  • Clean up $DATA
Batch job life cycle
  • Change directory to temporary dir ($WNTMP)
  • Copy files out of $DATA to $WNTMP
    • Done to ease the load on the NFS server
  • Run application executable
  • Tar up output files and copy to $DATA
  • Clean up working directory
More Information

Interoperability Notes / Questions

  • Does there exist something like $APP, $DATA, and $WNTMP for PRAGMA?