HPC/Applications/quantum-espresso: Difference between revisions

From CNM Wiki
Jump to navigation Jump to search
Line 11: Line 11:
* Wave function files from parallel runs are stored by MPI process number and are named sequentially <code>''prefix''.wfc''n''</code>. Subsequent calculations must use the same number of MPI processes, and each process must have access to its own ''n''. That means using a node-local directory is overly difficult.
* Wave function files from parallel runs are stored by MPI process number and are named sequentially <code>''prefix''.wfc''n''</code>. Subsequent calculations must use the same number of MPI processes, and each process must have access to its own ''n''. That means using a node-local directory is overly difficult.


I am reviewing the current sample.job file to work around these issues --[[User:Stern|stern]] ([[User talk:Stern|talk]]) 21:44, January 17, 2013 (CST)
<!--
<!--
With <code>''prefix''</code> being a user-specified name.
With <code>''prefix''</code> being a user-specified name.
http://www.quantum-espresso.org/wp-content/uploads/Doc/pw_user_guide/node9.html
http://www.quantum-espresso.org/wp-content/uploads/Doc/pw_user_guide/node9.html
-->
-->

Revision as of 22:09, January 18, 2013

Introduction

Quantum ESPRESSO is a suite of several programs working on the same data files. This brings with it some challenges for data management, and there are some problems.

  • A sample job file is at the customary location once the quantum-espresso module is loaded.
$QUANTUM_ESPRESSO_HOME/sample.job

Limitations

  • QE stores rather large wave function files in the directory specified by outdir or the ESPRESSO_TMPDIR environment variable. This may unduly tax the file server.
  • In a subdirectory prefix.save/ the main QE program pw.x stores important metadata needed for subsequent calculations such as a DOS calculation. This means the …TMP part in the name ESPRESSO_TMPDIR is misleading.
  • Wave function files from parallel runs are stored by MPI process number and are named sequentially prefix.wfcn. Subsequent calculations must use the same number of MPI processes, and each process must have access to its own n. That means using a node-local directory is overly difficult.