85th AMS Annual Meeting

Monday, 10 January 2005: 11:45 AM
Couple Distributed Earth System Models
Shujia Zhou, Northrop Grumman IT /TASC, Greenbelt, MD; and B. Womack and G. Higgins
Typical Earth system models consist of several complex model components coupled together through data exchanges. They require large amounts of data for initialization and may produce an even greater amount during execution. Ideally, an Earth system model would like to have the best combination of components available in the community and to execute on the platform that those components were optimized for. For validation, an Earth system model may also need to compare results using components from different institutions. However, efficiently distributing the data required by model components and managing their execution on multiple computer systems have been two major barriers to the development of distributed Earth system models.

In the last few years, several relevant technology advances have been made: Lambda networks, Grid Computing, Earth System Modeling Framework, and Common Component Architecture. Lambda networks provide transparent end-to-end dedicated bandwidth for fast delivery of huge amounts of data. A Lambda network is being constructed to connect institutions across the U.S. Grid computing integrates networking, communication, computation and information to provide a computation and data management capability that forms a virtual platform for information. Recently the Open Grid Services Architecture (OGSA) has been proposed to standardize Grid Computing. Grid-enabled applications are being demonstrated. The Earth System Modeling Framework (ESMF) is a community framework for Earth systems, funded by NASA’s ESTO/CT project. Its goal is to facilitate model coupling and to make models interoperable across organizations. The ESMF software consists of a superstructure (coupling) and an infrastructure (data structures and utilities). DOE also supports the Common Component Architecture (CCA) project defining a minimal set of standard interfaces for a high-performance component framework.

ESMF provides a component framework designed for the Earth system community. The current combination of increasing availability of high-speed dedicated bandwidth, standards for Grid applications, and high-performance component frameworks for Earth system models provide an important opportunity to evaluate how the ESMF architecture and its compliant applications can be extended to work in a Grid Computing environment. The current ESMF implementation is directed at running coupled model components on a local computer. Consequently, to couple ESMF compliant model components from different institutions, the source codes and their supporting environment such as input data and libraries have to be physically ported to one computer. Porting the source code for a complex Earth system model to a new machine is only the beginning. To be used operationally, the model must then go through a rigorous validation process to ensure that the porting process did not introduce any errors into the model. It may also need to undergo a performance optimization process to allow it to run efficiently on the new computer.

ESMF’s component design encapsulates an Earth system model component’s implementation behind a standard interface and provides a handle to access the component. This design facilitates interaction among model components and conforms in principle to the design of a Grid service. To explore this issue, we have Grid-enabled an ESMF-CCA Prototype 1.0 based on CCA’s Ccaffeine framework and the ESMF design. XCAT, a CCA compliant framework, is chosen to replace Ccaffeine since XCAT allows for CCA components to be compatible with Open Grid Standard Interface (OGSI) specification, which enables CCA components to be accessible via standard Grid clients. In this talk, we will describe the ESMF-CCA Prototype 2.0 and how its technology can be used to support Earth system models on the Grid. In addition, we will also discuss how a distributed CCA framework based on MPI (DCA) can be used to couple distributed, parallel Earth system models.

Supplementary URL: