Table of Contents

Subcommittee on Global Change Research, Participating Agencies and Executive Offices

Ad hoc Working Group on Climate Modeling

Foreword

Executive Summary

1. Background

2. Summary of Findings

3. Summary of recommen- dations

4. Final Comments

Charge to the Working Group 

Main Report

1. Purpose

2. Current Situation

3. Scope of Document / Underlying Definitions and Assumptions

4. Elements of Climate Science

5. Issues of Computational Systems

6. Human resources

7. Management / Business Practices / Institutional models

8. Recommen- dations

9. Reference Documents

10. Endnotes

Full Report (PDF)

1  For example, The Washington Post, June 12, 2000, pA03; June 14, 2000, pA38.

2  The term curiosity-driven is also possible.

3  To be clear, there are also high-end modeling activities that support discovery-driven research.  This is the traditional role of the Community Climate Model at NCAR.  This particular example does, in fact, have a product, the provision of a documented comprehensive model to the research community.  In this case a facility is being provided to the research community.

 4  http://www.ndsc.ncep.noaa.gov/

 5   i.e. massively parallel

 6  It is important to distinguish between computational capability and capacity.  Capability is the execution in a given wall-clock time of a job requiring the entire computational platform, and capacity is the aggregate of all the jobs running simultaneously on a platform.  Increasing capacity allows the execution of more jobs of the same size.  Increased capability is required to run larger jobs, e.g. more resolution, more comprehensive, or to achieve faster throughput of a same-size job.  Increased capability requires software development in order to utilize the potential capability of the hardware.  Increased capacity is easy to buy.  Increased capability is hard to build. 

7  Again, similar issues exist in the Data Element.

8  Further, a conscientious effort needs to be made to align resources and incentives with these efforts, as well as to separate the integration activities from short-term programmatic goals.  It must be recognized that the incentives to many individuals to advocate change to a more integrated climate service is low.  By in large, scientists have been successful and autonomous, and, by definition, movement to more product-oriented research removes some of their autonomy.  Furthermore, it is easy to point to numerous government-sponsored activities to 'centralize' computational systems, data activities, or modeling efforts that have either fizzled or failed.  Therefore, the trust in management is low.  Also, since the ultimate goal of an effort to integrate modeling activities will consume resources, some see a direct threat to their research activities as the product-generating model takes precedent over individual activities.

 9   i.e. massively parallel

 10   see Innovators Dilemma, in the reference list.

 11   see Ken Kennedy Testimony, in the reference list.

 12  There is little evidence that the market for high-end computing is adequate to motivate vendor development of easy-to-use systems software

 13  i.e. scalability is limited

14  A tremendous uncertainty in these arguments is the viability of the Japanese supercomputing industry.  Already delivered Japanese-made processors assure that U.S. scientists will remain behind for the next 3-5 years.  With the expected continued development and manufacturing of specialty Japanese processors, the gap will widen.   As long as U.S. computers are based on relatively slow processors in distributed memory architectures, and there are relatively fast Japanese processors with shared memory architectures, U.S. scientists will not catch up.  Increased capability will be much easier to achieve on the Japanese-processor based computers. 

 15   The environment remains "local" even if the university collaborator uses a computational facility at the collaborating center.

16   Science and Program Managers need to be fully aware that the bulk of the computational resources in a climate-science center go for testing and validation. 

 17   Controlled in both the scientific sense of controlled experimentation, as well as controlled in the sense of there being standard processes in the development of configured software suites for specific applications.

 18  see Ken Kennedy Testimony, in the reference list.

 19  see Top 500 web site.

 20  see Evaluating System Effectiveness in High Performance Computing Systems by AT Wong, L. Oliker, WTC Kramer, TL Kaltz and DH Bailey Lawrence Berkeley National Laboratory Draft Report 44542, November 11, 1999, available in PDF format.

 21  for example, The Washington Post, May 8, 2000, A21.

 22 This arises because the computing organization is often funded to pursue computational research within information technology programs or the computing facility is run as an institutional facility and the product generation exists in an uncomfortable balance with large numbers of small discovery-driven research projects. 

23  This focus on individuality manifests itself in large projects through the development of consensus decision making.  It becomes necessary for all members of the project to have ownership of the project.  Therefore, such projects are generally driven by the development of individual capabilities with only collegial or casual integration towards project goals.  The needed infrastructure, including process definition, is generally underdetermined and under funded.  This leads to the project taxing the, generally, already stressed schedules and resources of the individual participates, who are generally not directly rewarded for their participation in the project.

 24 If the decision is made to locate the Climate Service in an existing Agency or center, then the metrics of success of the Agency or center must be realigned with the success of the Climate Service a critical metric of Agency or center success. Agency mission and business practices would need to be altered to support the delivery of the products of the Climate Service.

25 Operational capabilities rely on data access.  There is much in common with climate and weather data, and shared data infrastructure would seem to be at the core of any and all operational capabilities.