Grid Computing: The Development of a Portal
for Model Test and Verification
Mark
W. Govett
Forecast Systems Laboratory, OAR
Testing has begun on two variants of the Weather Research and
Forecast (WRF) model to determine the best candidates that will
be part of a six member ensemble slated to become the operational
weather prediction model for the National Weather Service in October
2004. A testing procedure has been identified in the WRF test
plan that will require 1255 tests to compare the two variants.
Many other WRF variants and model ensembles have been proposed
for future comparisons that will likely require thousands of additional
tests. The management of software, data and tests will require
significant resources at multiple super-computing sites both during
initial WRF testing and for successive model research and development.
To simplify this process, the Forecast Systems Laboratory has
begun development of a WRF portal that will be used for the test,
evaluation and verification of changes made to the WRF model.
A
portal is broadly defined to be a gateway in which users can access
and utilize distributed compute resources (eg. systems, data,
network) without regard to what or where they are. Both grid and
web technologies will be used in the development of a WRF portal.
Grid technologies will be employed to provide a powerful and flexible
way to access and combine resources from multiple institutions
so that tests can be run or data results can be obtained more
efficiently. Web technologies will be used to simplify access
to the WRF model for testing and evaluation by the larger meteorological
community. The portal will incorporate all the elements of end-to-end
model testing including initialization, model forecast , post-processing,
and verification. In designing the portal, accessibility, extensibility,
software management, data management, user interface, and security
issues will be considered. This talk will describe the WRF portal
concept and discuss issues relevant to grid computing, web services
and the design of a broad-based modeling infrastructure.
