---

*  A M R 2 0 0 3    B E N C H M A R K       S E S S I O N *

Chicago Workshop on Adaptive Mesh Refinement Methods

hosted by the ASCI/Alliances Flash Center
for Astrophysical Thermonuclear Flashes
University of Chicago

Workshop to be held in Chicago, September 3-5, 2003

** Special Session to be held on September 5, 2003 **

---

SUMMARY

The Initial Call for Participation included problems that could be characterized as follows:

• (A) AMR kernel (physics independent);
• (B) single physics (hyperbolic, elliptic, others);
• (C) multiphysics.

There seems to be a relative consensus regarding class A and B benchmarks, and we decided to use them for comparison between different implementations.

Class C benchmarks will be included in the final presentation for demonstration purposes and to allow for future comparisons; you are encouraged to present your most favourite problem. Please keep in mind that the test cases should not depend on data or parameters which are not publicly available - such dependencies will make your test less valuable for other reserchers and a weaker candidate for publication in the proceedings.

BENCHMARK CHARACTERISTICS

1. Each benchmark problem should have well-defined initial and boundary conditions and an unambiguous solution (to allow for self-validation).
2. The purpose of the test, and an easily measured metric of performance, should be specified.
3. The problem size should be scalable (through refinement) to run on machines from high-end desktop machines to supercomputers.

BENCHMARK PROBLEMS

Please review the following information and contact us if it is insufficient to set up and run your model.

•  
• AMR kernel benchmark
•  
  • static structure
  • represent a thin shell on a 3-D cartesian mesh
  • record the total numer of cells used (not including any ghost cells)

    mesh geometry	cartesian
    box size	[0.0, 1.0]
    shell inner radius	0.5
    shell outer radius	0.75
    effective resolutions	128^3, 256^3, 512^3, 1024^3

  
  Notes
  1. Original AR kernel benchmark considers one octant of the sphere. Possible variations include half and full sphere - this may help us to identify certain inconsistencies in the regridding kernel.
  2. Refinement criterion: refine the shell edges to the highest resolution.
  3. If implementation uses "grid efficiency": (a) adopt 70%, (b) use the most optimal value.
  4. If implementation uses "buffer cells": (a) use none, (b) use 2 cells wide buffer.
  5. If implementation uses arbitrary "refinement ratio": (a) use r=2, (b) use optimal r.
•  
• Hyperbolic benchmark (single physics)
•  
  • compressible Euler equations, gamma-law gas
  • Sedov point-like explosion on a 3-D cartesian mesh
  • record time to solution

    box size	[-1.0, +1.0]
    density (ambient)	1.0
    pressure (ambient)	1.e-5
    gamma	1.4
    final time	0.05
    initial explosion radius	4 cells
    explosion energy	1.0
    boundary conditions	transmitting (outflow, zero gradient)
    effective resolutions	128^3, 256^3, 512^3, 1024^3

  
  Notes
  1. Refinement criterion: (density jump, pressure jump) < 0.1. Jumps are defined as max(a(i-1),a(i+1))/min(a(i-1),a(i+1))-1.
  2. If implementation uses "grid efficiency" adopt the optimal value.
  3. The explosion is initiated by increasing internal gas energy inside a small spherewith radius of 4 cells at the finest resolution and centered on the grid origin. Use the best method to deposit energy and make the sphere "rounded".

BENCHMARK SYSTEM

We will have 3 weeks of access to the brand new shared memory SGI Altix 3000 server generously donated to the workshop by SGI. We expect the machine will have at least 12 CPUs (Intel Itanium-2) and 2 Gbytes per processor. The system runs a custom 64-bit version of the Linux OS and uses Intel 7.x compilers. More details will be provided soon.

Participants of the benchmark session will be granted free access to the system starting about 10 days before beginning of the workshop (this should allow for porting your code to the Altix platform). We cannot guarantee availability of the system later than about a week after the workshop.

It is essential you inform us about your wish to participate in the benchmarking session by sending email to amr2003@flash.uchicago.edu before August 23. This will ensure you being granted with maximum access time to the system.

BENCHMARK SESSION

There will be two 2-hour long sessions on the morning of September 5 with time equally split between all session participants. Each session presentation will include a short description of the code followed by the benchmark results.

BENCHMARK PUBLICATION

Benchmark session results will be presented in form of regular article to be included in the workshop proceedings.

CONTACT INFORMATION

For any questions regarding the benchmark session email the local organizing committee at loc-amr2003@flash.uchicago.edu

GENERAL INQUIRIES
 

Carrie Eder
Chicago AMR Workshop
The ASCI Flash Center	www: http://flash.uchicago.edu/amr2003
The University of Chicago	email: amr2003@flash.uchicago.edu
5640 S. Ellis, RI 468	ph: +1-773-834-2057
Chicago, IL 60637, USA	fax: +1-773-834-3230

---

Last modified:

The ASCI Flash Center is based at the University of Chicago under U.S. Department of Energy contract B341495. All material on these pages is Copyright © 2003 ASCI Flash Center. Please report problems with this page to: webmaster@flash.uchicago.edu