NERSC logo National Energy Research Scientific Computing Center
  A DOE Office of Science User Facility
  at Lawrence Berkeley National Laboratory

Help for the ERCAP Code Description Questions

The code descriptions you provide are used by DOE and NERSC to evaluate the type and extent of usage of machine resources that users require. Also, DOE's COmputational Review Panel will provide a computational rating for your project based on these answers.


Help for the 2005 ERCAP Code Description Questions

19.1 Code and Application Descriptions

A code description should be provided for each of your project's most frequently used codes and applications. The information for each code is contained under a tab for that code. Click on the "Add another code" tab to type in your answers for another code.

Code Name: This is the name by which your code is known, e.g. NIMROD, Gaussian.

Code Description: Provide a brief description of the code.What does the code do? What is its underlying mathematical formulation? What are the numerical techniques employed?

Machines to be used: On which DOE Office of Science machines do you plan to run this code (assuming you get an award)? What percent of the project's overall allocation do you expect this code to use on the machines you list?

Code Algorithms: Check as many algorithms as apply from the checklist. You can enter additional algorithms in the Other box.

Languages Used: Check as many languages and communications libraries as apply from the checklist. You can enter additional languages or communications libraries in the Other box.

Libraries Called: Check as many libraries as apply from the checklist. You can enter additional libraries in the Other box. Note that communications libraries (like MPI) are listed under Languages Used.

Limits to Performance: Check as many limits to performance as apply from the checklist. You can enter comments about limits to performance in the box. You can enter additional (brief) limits to performance in the Other box. You can use poe+ to help identify the limits to performance of Seaborg codes. If you need additional help, contact a consultant at consult@nersc.gov.

Performance Comments: You can comment on the limits to performance, checked above, or on performance results you enter in the 19.2 Code and Application Performance table.

Does the code checkpoint? Select Yes or No.

Other Software Used: Describe other software used in conjunction with this code: applications for pre and post processing, visualization, job control, etc.

Plans for Code Development or Enhancement: Briefly describe your plans.

19.2 Code and Application Performance

Provide code performance data for typical processor counts used in production in 2004 (or expected for 2005 production). For machines with more than one processor per node enter # of procs as the number of nodes used times the number of processors per node.

To answer this question for Seaborg codes you need to be familiar with collecting hardware counter data on the IBM SP. Documentation on using hpmcount and poe+ is provided at http://www.nersc.gov/nusers/resources/software/ibm/hpmcount/poe+.php.

It is recommended that you use NERSC's poe+ command. Two new lines have been added to the poe+ report that allow you to easily fill in the code performance table: ERCAP GFLOP/S and ERCAP MEMORY. Memory is the aggregate high water memory (in gigabytes) used on that number of processors.

In general you need not report more than several data points per code. You can use as many rows as needed in this table: the form will automatically increment the number of available table rows by 2 after you fill in and save the currently available number of rows.

You may provide brief comments for these numbers in the How info was collected / comments box. Lengthier comments can be entered in the Performance Comments box in question 19.1.

Note: Enter only numbers in the # of Procs, Total Gflops and Memory columns.


Help for the 2004 ERCAP Code Description Questions

16.2(a) Seaborg code performance and scalability

To answer this question you need to be familiar with collecting hardware counter data on the IBM SP. Documentation on using hpmcount and poe+ is provided at http://www.nersc.gov/nusers/resources/software/ibm/hpmcount/poe+.php.

It is recommended that you use NERSC's poe+ command. Two new lines have been added to the poe+ report that allow you to easily fill in the code performance table: ERCAP GFLOP/S and ERCAP MEMORY.

For each of your major codes, please fill in the code performance table for the following processor counts:

  1. Typical processor counts that you plan to use for production runs in FY 2004.
  2. The largest processor count for which you have performance data.
  3. If you wish, a few intermediary rows to show scaling data.
In general you need not report more than several data points per code. You can use as many rows as needed in this table: the form will automatically increment the number of available table rows by 2 after you fill in and save the currently available number of rows.

You may provide comments for these numbers in question 16.2 (b).

Notes on entering table values:

  1. You cannot enter only a code name without also entering the number of processors. If you enter only a code name and hit save that row will not be saved.

  2. No matter the order you enter them, the rows are displayed sorted alphabetically by code name and then by increasing number of processors.


LBNL Home
Page last modified: Mon, 16 Aug 2004 19:36:00 GMT
Page URL: http://www.nersc.gov/nusers/accounts/allocations/ercap/codehelp.php
Web contact: webmaster@nersc.gov
Computing questions: consult@nersc.gov

Privacy and Security Notice
DOE Office of Science