On Thursday, January 14, 2016, the STORM team presented several talks during the Second Symposium on High Performance Computing for Weather, Water, and Climate at the American Meteorological Society 96th Annual Meeting.
Click on the links below to watch the recorded presentations:
On December 10, 2015, Zach Byerly presented a paper outlining a technique for interfacing C and C++ codes to Fortran subroutines. This was a direct result of the STORM team’s efforts to join HPX and LibGeoDecomp to the discontinuous Galerkin version of ADCIRC, DGSWEM.
Many HPC applications developed over the past two
decades have used Fortran and MPI-based parallelization. As
the size of today’s HPC resources continues to increase, these
codes struggle to efficiently utilize the million-way parallelism of
these platforms. Rewriting these codes from scratch to leverage
modern programming paradigms would be time-consuming and
error-prone. We evaluate a robust approach for interfacing
with next-generation C++-based libraries and drivers. We have
successfully used this technique to modify the Fortran code
DGSWEM (Discontinuous Galerkin Shallow Water Equation
Model), allowing it to take advantage of the new parallel runtime
system HPX. Our goal was to make as few modifications to the
DGSWEM Fortran source code as possible, thereby minimizing
the chances of introducing bugs and reducing the amount of
re-verification that needed to be done.
The 2015 ADCIRC Users’ Group Meeting was held March 30th and 31st at NOAA’s Center for Weather and Climate Prediction. The facility (both inside and outside) was beautiful, and Jesse Feyen did a fantastic job organizing and hosting the event. Several members of the STORM team were in attendance.
Over the Mardi Gras break, the STORM project was represented at the 2015 SI2 PI workshop on software sustainability. In order to prepare for the workshop, we spent some time outlining the motivation, goals, and impacts the at this point of time in the project as well as graphically displaying how all the different components of the project are inter related. We summarized these concepts on a poster which captures the highlights of our work. Motivation: We intend to reinvigorate ADCIRC by leveraging the scalability harnessed by HPX, an open source C++ runtime system, and improving the algorithms which provide the foundation of the model.
Scale to 256K cores on modern HPC Systems
Separate software infrastructure from application code
Use multiple solution algorithms for increased accuracy and fidelity
Impact: STORM will have a large impact on the communities that it comes in contact with including reseachers who study coastal modeling, discontinuous Galerkin methods, and unstructured grids. Consumers of ADCIRC hydrodynamic output such as biologist, ecologists, coastal engineers, and emergency managers will benefit from the improved resolution and fidelity built into the code base. Finally, this project will introduce daily simulations of Louisiana coast lines which will be made available to communities such as fishermen and wetland researchers who could benefit from this information.
Community Engagement: This diagram, which illustrate how several key parts of the project interact with different relevant communities:
We have developed a “proxy app” which emulates the communication patterns of ADCIRC. gameoflife_adcirc uses LibGeoDecomp, an auto-parallelizing stencil library, and HPX, a next-generation parallel runtime system, to manage parallelism and boundary exchange. A decomposed ADCIRC computational mesh is used as the basis for a very simple “game of life” style cellular automaton kernel written in FORTRAN. This proxy code is the first step in determining the best path for a next-generation ADCIRC. These results were presented at Supercomputing 2014 in New Orleans, LA at the Intel Parallel Universe Theater. The visualization below is generated with an early version of the CERA viewer app, developed by Carola Kaiser.
The STORM team held its kickoff meeting at LSU’s CCT on October 21st and 22nd. At the meeting, each participating institution gave presentations on the research that they will be focusing on during the duration of the project. Amongst the topics covered was the proposed software stack, research on DG methods, and improvements being made to the CERA viewer. If you are interested in viewing the presentations yourself, you can find them on the kickoff meeting event page.