The Geographic Environmental Modeling System (GEMS)

Contents

Framework Overview

The GEMS framework consists of five components, the User Interface provides a consistent interface to all the system capabilities. The Execution component provides control over air quality model calculations - the user may choose from several different scientific models and the diverse computational platforms available. The Visualization component provides advanced capabilities for viewing the data output from the air quality model calculations. The Data Management component manages the data produced and consumed by all components of the system. The Monitoring component provide capabilities for tracking performance and faults in the distributed system on which all the components operate.

The Five Components of the GEMS Framework and the Current Prototype Implementation

User Interface

The GEMS user interface provides capabilities for analysis of the data generated by the underlying air quality models. Data from several different control strategy scenarios can be compared in order to find the "best" strategy. The picture below shows the interface while comparing two sets of data from Los Angeles simulations. The first pane (upper left) shows the data for the basecase measurements. The second pane contains data from a modified run of the program done as part of a sensitivity study - several parameters in the chemical equations were varied to determine the sensitivity of ozone concentrations to these types of perturbations. The third pane contains a deficit enhancement plot showing the absolute difference between the data in the first and second panes.
Control Strategy Comparison

Execution

The Execution component allows the user to choose the most appropriate computational model and computing platform to use for a given simulation. Current model choices include the single-scale CIT Airshed Model, the multiscale Urban and Regional Model, and the EPA's Urban Airshed Model. Computing platforms available include single Alpha or SPARC workstations, a SuperCluster (or "farm") of Alpha workstations connected with a high-speed network, the Cray C90 vector supercomputer, or the Cray T3D massively-parallel machine. Parts of the URM system are also available on the Intel Paragon and Intel iWarp parallel platforms.
Choice of Models and Platforms

Visualization

The Visualization component provides advanced capabilities for viewing the data generated by the air quality models. The picture below shows an isosurface enclosing the volume of air where the concentration of ozone exceeds the National Ambient Air Quality Standard (NAAQS) of 0.12 ppm. The map below the ozone cloud contains both LANDSAT satellite imagery and political boundaries derived from the Census Department's TIGER database.
Ozone Cloud over Los Angeles

Data Management

The Data Management component is responsible for all data stored by GEMS and the air quality models. The system must provide support for a variety of datatypes, storage sub-systems, and access patterns. Different components of the system will be written in different languages (currently - Fortran, C, and C++) and will run on a variety of heterogeneous platforms. The data will also be physically distributed on servers across the country. The Data Management component must provide the appropriate data conversions between different formats, provide location-transparent access for all clients, and provide adequate performance (I/O operations should never be the bottleneck in the system).
Distributed Data Servers and Clients

Monitoring

The Monitoring component allows the system and the user to keep track of what is going on across the distributed system of machines that constitute the GEMS system. Monitoring capabilities are used to identify performance bottlenecks in the system, and provide information about network and system faults as they occur. The picture below shows output from the BEE monitoring system that is currently being used with GEMS. The display shows the variation in the load on several processors cooperating on a URM mutliscale model calculation. The second picture gives some idea of the highly distributed nature of the GEMS system. There may be modelers on the CMU campus and at the EPA in North Carolina both making use of resources at the PSC in Pittsburgh. The data used for different evaluations might be stored at any of these three sites. In the future, it should be possible for the modelers not only to share data within the system, but also to collaborate interactively using the same interface.
The BEE Monitoring System
CMU/HPCC/GEMS/Version 1.0/12-Nov-94/er1p