Student research opportunities
Advanced Parallelization Techniques and Supporting Infrastructure for large-scale Climate Science Simulations
Project Code: CECS_629
This project is available at the following levels:
CS single semester, Honours, Summer Scholar, Masters, PhD
Keywords:
Parallel computing, climate change science, climate simulation
Supervisor:
Dr Peter StrazdinsOutline:
Climate and weather models currently consume vast amounts of supercomputer time, with the most dominant component being the atmosphere. In 2011 and beyond, the plans of BoM
to make longer-range forecasts by increasing model resolution, and the ACCESS project performing large-scale climate simulations, the amount of usage is set to explode.
However, these models are complex software systems, with large amounts of legacy code. The primary consideration is to correctly encode the science for meaningful simulations; the secondary is performance, particularly on large-scale parallel computers. As a result, generally only the simplest parallelization strategies are reflected in current software. This project will explore new techniques for the optimization and parallelization of current climate model codes, including the exploration of new programming paradigms.
A second (possibly alternate) theme to the project would be to develop supporting infrastructure in order to optimize such codes reliably. This is non-trivial due to the complexity and the long history of such large-scale numerical simulations.
At the Honours or Masters level, specific issues on the MetUM (UK Met Office Unified (atmosphere) Model) will be investigated. These could include code structure and performance analysis (with tuning) of:
* I/O process. The models needs to regularly dump large amounts of generated data during a simulation. This is the major bottleneck to scalability
* Improving the memory performance of the computationally dominant part of the model, the acoustic wave solver.
* Investigating the effect of mixed OpenMP/MPI parallelism. This offers a potential advantage over the MPI-only implementation, as it reduces the volume of communication of grid points.
Goals of this project
The goals of this project include (1) analyzing and developing an understanding of the performance and scaling behaviour of a selection of climate model codes, (2) exploring and evaluating new opportunities and techniques for parallelization. This includes the use of new programming paradigms; of particular interest will be performance issues on systems made up from manycore chips.
Requirements/Prerequisites
(PhD Level) An Honours degree in computer or computational science or equivalent. Some background in high performance computing and mathematical modelling is desirable.
Work on supporting infrastructure includes automated methods to reverse-engineer test harnesses (correctness and performance) for selected performance-critical subroutines.
Student Gain
Climate science is of increasing importance, and the with it the need to perform efficient and meaningful simulations, especially for medium to long term forecasts. This project represents an opportunity to join and make a significant contribution a national team working in climate change science, including CSIRO, Bureau of Meteorology and the ACCESS project.
Links
The Community Earth System ModelThe Vayu cluster
Prelinary Experiences on the UM
