15-May-2008

Introduction

Running the Espresso pw.x program requires consideration of the amount of memory that the job will require with specific data input. Historically pw.x has exceeded the amount of memory available on RTC's compute nodes (4GB per node). When the job exceeds 4GB of memory, the node will likely crash. Many jobs of this type submitted at the same time can result in a large number of crashed nodes. Below is a reference to documentation that will describe how to avoid this problem.

Quantum-ESPRESSO Users Guide

The Quantum-ESPRESSO Users Guide includes information on how to determine the memory requirements of a pw.x job with specific data input. On page 49 of the 3.1.1 manual, it explicitly mentions a routine in the code, memory.x which gives you an estimate of memory required to run the job, given the same inputs as the pw.x program.

All pw.x users should use memory.x to determine the amount of memory needed by their jobs. If this amount is less than 4GB, then it is okay to submit these jobs to the RTC compute nodes but it is recommended that you request exclusive access so that you will have access to all 4GB rather than sharing the memory with someone else's job. If this amount exceeds 4GB, then it will be necessary to run the jobs on one of the large memory nodes available in the super queue. Information on both topics can be found in our FAQ.

NOTE: Do not run pw.x on a login node. These nodes are shared by all users on the RTC and are not intended for compute jobs. Any compute job running on the login nodes for more than 30 minutes is subject to termination without warning!!