The GFDL Flexible Modeling System Runtime Environment

Amy Langenhorst
Geophysical Fluid Dynamics Laboratory, OAR

The GFDL Flexible Modeling System (FMS), presented at NOAATech 2002, is a scalable framework for constructing climate models developed at NOAA's Geophysical Fluid Dynamics Laboratory. The framework permits a distributed development model, with independent teams of researchers able to contribute codes for different components of the climate system. Assembling these into a whole can be a daunting task. The FMS established standards for constructing climate model components expressed in high-level data structures; provided a shared technical infrastructure for operations common to many model components, such as parallelism, I/O and output of model diagnostics; and supplied a standard mechanism for coupling and exchanging data between model components on independent grids and parallel composition. The FMS permitted scientists to construct models easily from a hierarchy of choices of increasing complexity from available components, ranging from idealized studies to comprehensive general circulation models for full-scale assessments. Climate models built upon the FMS have been in production at GFDL since 2001, and scientific results are already in the peer-reviewed literature.

A production environment requires a set of software tools around a complex code like FMS. The FMS Runtime Environment (FRE) provides a software infrastructure and user interface for FMS. The FRE covers the full sequence of user operations required to run FMS-based models on complex hardware platforms. Capabilities provided by the FMS Runtime Environment include:

- source code management: revision control utilizing CVS, assembly of model code out of a wide choice of available components tailored to scientific needs for particular experiments;

- model configuration: configuration of model components, setting of scientific input parameters and datasets;

- platform configuration: settings optimizing the code to the runtime hardware platform;
- job control: management of very long runs accomplished over multiple submissions; resource allocation and scheduling;

- model output management: postprocessing of output data to assemble datasets suitable for analysis;

- extensible standardized diagnostic analysis suites: batch software to configure and execute analysis of model output, making graphical results available over the web in standard format;

- experiment database: storage, retrieval and comparison of different experiment configurations through a web SQL interface.

The entire FMS Runtime Environment is expressed in XML. Scientists edit an XML model description file to configure and customize each step of the production process: code extraction, compilation, execution, postprocessing and analysis. A key feature of this approach is that the XML provides a standard mechanism for a comprehensive description of model configurations. This can be layered on top of the metadata conventions for model fields (such as CF) to provide a standard description not only of the variables within a model output dataset, but also the model itself: information about model grids, dynamics and physics choices, and all scientific input parameters and datasets. Command-line utilities act upon the XML, and a GUI is planned.

The talk will cover the rationale for the FMS Runtime Environment, an overview of the XML syntax with examples of simple and comprehensive models, and future directions for high-level descriptions of model output datasets.