SUBJECT: Proposed Interaction of the DOE-funded SciDAC CCSM Consortium project with the NASA-funded Earth System Modeling Framework project
Background
The SciDAC CCSM Consortium project and the Earth System Modeling Framework (ESMF) project are two closely allied efforts involving improvements to the Community Climate System Model (CCSM). The five-year SciDAC project, funded by DOE in October 2001, will improve both the computational and scientific capabilities of the CCSM. The three-year Earth System Modeling Framework (ESMF) project, funded by NASA in March 2002, will create a common modeling infrastructure that the Earth science community can leverage to solve common computational problems. The CCSM is an ESMF testbed application.
These two projects intersect primarily in the coupling system for the CCSM. The precursor to the SciDAC project, the eighteen-month DOE Accelerated Climate Prediction Initiative (ACPI) "Avant Garde" effort, had as one of its focus areas the development of a new CCSM coupler, CPL6. The current CCSM coupler, CPL5, is not distributed and would become a performance bottleneck for the CCSM as the modeling effort grows. CPL6 development began under ACPI and continues under SciDAC. One of the important outgrowths of this effort is the creation of the Model Coupling Toolkit (MCT) by Jay Larson and Rob Jacob of Argonne National Laboratory (ANL). The concepts and implementation of MCT are of great potential benefit to the ESMF effort, which is also charged with developing flexible coupling tools to increase interoperability and reuse across a large set of climate and weather models and data assimilation systems. Another important contribution to ESMF will be the Spherical Coordinate Remapping and Interpolation Package (SCRIP) developed by Phil Jones of Los Alamos National Laboratory (LANL). SCRIP is used in CPL5 and will be used, in modified form, in CPL6 and ESMF.
Participation of DOE scientists in ESMF
It is critical that a computationally improved CCSM, including a coupler with both performance and flexibility enhancements, be delivered in a timely manner to meet the needs of both CCSM scientists and DOE climate program objectives. The DOE SciDAC CCSM Consortium project has a commitment to support CCSM development. The CCSM project, in turn, is committed to evaluating the ESMF software, including ESMF coupling services, and to adopting the software if it meets project requirements for performance, robustness, ease of use, and flexibility. To maximize the benefit to ESMF of DOE work on the CCSM - specifically MCT and SCRIP - it is important to involve DOE scientists in the definition of requirements for and early prototyping of ESMF. In order that such contributions to ESMF do not adversely impact the progress of DOE-funded SciDAC activities, participation of DOE scientists in ESMF must be supported by NASA funding.
Current Status and Plans
In addition to developing MCT, which includes tools for representing and transferring gridded data and its associated metadata, Larson and Jacob advised an NCAR CGD team, led by Tom Bettge, on the development of higher-level objects such as bundles of fields and distributed grids. A unit tester of CPL6, called YAK, has been implemented, and it successfully transfers data between models and the coupler. The prototype coupler is distributed, has shown some performance enhancement over CPL5, and greatly increases flexibility in modifying the fields sent between models. CPL6 requires some functional enhancement, performance optimization, and improvements in robustness before it becomes viable as a replacement for CPL5.
The functionality represented by MCT will be necessary for ESMF, and, like other coupling software such as the Flexible Modeling System from GFDL and the Goddard Earth Modeling System, will likely come close to satisfying ESMF requirements. As such it is an excellent prototype for ESMF coupling services. Like Larson, Phil Jones is a joint SciDAC/ESMF investigator, whose SCRIP software will be optimized and extended under ESMF funding for use in the framework. Will Sawyer of the NASA Goddard Data Assimilation Office is also working on both the SciDAC and ESMF project, contributing his experience in developing the PILGRIM (Parallel Integrated Library for GRId Manipulation) communication library. However, the scope of ESMF also encompasses extension and optimization of data types for time-variant, ungridded observations; introduction into the coupler of on-line generation of grid-interpolation weights; additional high-level features such as a component registry; sets of utilities unrelated to coupling; and, perhaps most importantly, the collective standardization of interfaces so that the modeling community can more easily interchange models and reuse coupling and other infrastructure software.
Currently the ESMF team is beginning a requirements collection process that will extend through early summer 2002. A specification and prototype implementation are scheduled for summer 2003. A production version of the ESMF is anticipated in summer 2004. If ESMF meets the requirements of the CCSM, it will be adopted in the production version of CCSM as scheduling permits.
Although currently unfunded by NASA, Larson is an investigator on the ESMF project and has been active in meetings and teleconferences since the ESMF collaboration formed in early 2000. One of the critical roles that the Argonne team has played is as a liaison to the broader computer science community for both the CCSM and ESMF. Larson is a collaborator on the Common Component Architecture (CCA) project, a DOE effort that seeks to enhance interoperability of scientific applications through CORBA-like standardization of components and data transfer services. He has been conducting research into alternative programming languages for coupling tools and into language interoperability tools. These areas are of central importance to the ESMF, which has committed to providing multiple language bindings. Through Larson, the ESMF and CCSM have been informed of new computer science opportunities and results, and the ESMF and CCSM have been represented to the computer science community.
Continued development of MCT
Because the purpose of the SciDAC CCSM Consortium is to support development of CCSM, the Consortium management position is to endorse ESMF to the extent that it is approved by the Scientific Steering Committee, the ultimate decision-making body of the CCSM. It is hoped and expected that the expertise and experience acquired by DOE scientists during the development of MCT and SCRIP inform ESMF. Continued development of MCT in the near term
(~1-2 years) is necessary to meet SciDAC goals, and should not be construed as competing with ESMF. This is consistent with CCSM plans to incorporate a completed and proven CPL6 into the CCSM during 2002. In the longer term, SciDAC support for MCT may continue under the following conditions:
Of course, MCT development will continue, regardless of the "Consortium", if MCT has another (paying) customer whose needs are best met by MCT.
Conclusions
It is essential for the alignment of these projects and a smooth transition of the CCSM from MCT-based couplers to ESMF-based couplers that Larson and Jacob participate in the earliest stages of ESMF requirements definition, interface specification, and prototyping. To maintain strong ties with the computer science activities at Argonne and elsewhere in the DOE, it is also critical for Larson to maintain a continuing role as a computer science liaison to the ESMF and CCSM software efforts. Therefore, Larson and Jacob will continue developing, extending, and optimizing coupling services for the CCSM throughout the 5-year course of the SciDAC project. The focus of their efforts will shift from a MCT-based coupler to an ESMF-based coupler once ESMF is installed in a production version of the CCSM. In order that such contributions to ESMF do not adversely impact the progress of DOE-funded SciDAC activities, participation of DOE scientists in ESMF must be supported by NASA funding.
Endorsed for SciDAC by:
David Bader, DOE, SciDAC Program Manager
Robert Malone, Los Alamos National Laboratory, Co-PI SciDAC Consortium
John Drake, Oak Ridge National Laboratory, Co-PI SciDAC Consortium
John Taylor, Argonne National Laboratory, Co-I SciDAC Consortium
Endorsed
for ESMF by:
Tim Killeen, National Center for Atmospheric Research, Co-PI ESMF
Arlindo da Silva, NASA GSFC Data Assimilation Office, Co-PI ESMF
John Marshall, MIT, Co-PI ESMF
Cecelia DeLuca, National Center for Atmospheric Research, Co-I, ESMF