About Parallel Programming Services
We are a professional group of physical scientists with backgrounds (to Ph.D. Level) in physics, chemistry and meteorology, distributed around Europe and led by Enda O'Brien and Alastair McKinstry from Galway, Ireland.
We have proven expertise in parallel programming and many years experience in the high-performance computing industry.
Our accomplishments include:
- Speeding up the performance of an operational weather model from Canada by a factor of 7.
- Optimizing the message-passing in a European operational weather model to allow improved scaling, effectively doubling the model's performance on 256 processors.
- Enhancing a global weather model from Asia, which initially had just a 1-dimensional domain decomposition (in latitude), by adding a second dimension (either in longitude or in the vertical, depending on whether the application was in grid-space or spectral-space) to the domain decomposition. This enhanced domain decomposition means that the model can now run on potentially thousands of processors, rather than a smaller number (two or three-hundred) determined by the number of model grid-points in the meridional direction.
- Implementing full parallelisation (in both grid-space and spectral space) in a weather model from eastern Europe, which was initially parallelised in grid-space only. Parallel speedup benefited hugely, in accordance with Amdahl's law.
In a nutshell, we took applications whose developers had started the parallelization process, and we completed the job.