MAPP.html

Traditionally, NOAA HPC systems have been managed individually. In an effort to optimize the use of its High Performance Computer (HPC) systems and reduce costs, NOAA is combining the procurement of its three HPC development systems at the Geophysical Fluid Dynamics Laboratory (GFDL), the National Centers for Environmental Prediction (NCEP) and the Forecast Systems Laboratory (FSL) into a single procurement. This reflects NOAA's view of its HPC systems as an enterprise resource; where each HPC system will be available to any NOAA users who require them. The joint procurement is currently under way with a contract award anticipated in September 2005.

As NOAA stresses the increasing importance of sharing Information Technology (IT) resources across the organization, code interoperability across HPC systems at the FSL, GFDL, and NCEP has become a critical requirement. Code interoperability has immediate benefit to NOAA and is implicit in the joint procurement. It's not enough to simply achieve interoperability, however; improved usability of these codes is also required. To improve usability a supporting computing environment must be provided to get users across NOAA to use the ported codes and to utilize remote system resources to run tests, locate files, transfer files, and verify modeling results. The MAPP project, funded through NOAA's High Performance Computing and Communications (HPCC) office, is intended to meet both the requirements that codes be portable across HPC systems, and that users can easily run them on remote resources from their desktop systems.

The two primary objectives of the MAPP project are: (1) to port primary scientific codes to NOAA's three HPC systems, and (2) to develop a web-based portal to permit users to configure and run the codes on these systems. These objectives will be further described. Project milestones and deliverables are also described in the MAPP milestones document.

A. Porting Scientific Codes

FSL will lead the effort to port primary applications to each of the HPC systems at FSL, GFDL, and NCEP. FSL will performance tune these applications as necessary and provide capabilities for software source control and to develop run-time scripts and regression tests in order to insure correctness. This work is similar to how FSL's Advanced Computing Branch (ACB) currently handles portability and performance of scientific codes it has worked with. Based on guidance from researchers at GFDL, NCEP and FSL, we have selected three codes to be ported, and performance tuned. These models, including requisite pre and post processing software, are listed:

B. Web-based Portal Development

FSL will also develop a Web-based Portal with common programming interfaces to permit applications to be run on all of NOAA's HPC systems. This work will build on recent efforts to develop a prototype portal in which to run WRF for the Developmental Testbed Center (DTC). A portal is a web-based application designed to connect users with distributed resources necessary to accomplish desired tasks. This portal will provide an environment to support model development, testing and verification. The portal will be a java-based web application that will allow users to access remote resources to create model domains, configure model runs, run the models, monitoring the tasks while they are running, and inform the users when they have completed. An early release of the portal will be available here by the summer of 2005.

The ACB also plans to use non MAPP funds to extend the portal to further research goals in exploring and applying web and grid technologies. For example, we want to develop capabilities to discover and transfer remote files necessary to initialize model runs. This will reduce or eliminate the requirement that data be stored locally, and reduce the need for duplicate data repositories that must be supported, migrated, and managed. We also want to enhance search capabilities to permit users to search for previously run tests - thus eliminating or reducing the need to run cases that have already been run. Finally, we want to integrate a NOAA meta-scheduler into the portal to permit users to run tasks on a NOAA HPC system without worrying about where they actually run. A meta-scheduler could increase systems utilization, improve job throughput by balancing job loads between HPC systems, and lead to reduced costs.