EPCC (www.epcc.ed.ac.uk) is the supercomputer centre at the University of Edinburgh, and one of the leading High Performance Computing (HPC) centres in Europe. This PhD studentship is funded and co-supervised by AWE (www.awe.co.uk). Owing to the nature of the work done at AWE only British Nationals are eligible for the studentship, and the student will be expected to undergo security clearance.
Scalability and Performance of Multi-Physics Codes on Petaflops Parallel Supercomputers
Multi-physics codes are defined as those which employ more than one significant area of physics in the same code. The particular focus of interest in this PhD studentship is the case where both implicit and explicit solution techniques are required in the same code. An example would be an astrophysics simulation using explicit methods (eg direct time integration) for hydrodynamics computation and implicit methods (requiring the solution of a system of linear equations) for radiation transport. There are particular challenges in enabling such codes to run efficiently on the world's largest parallel supercomputers.
The code "Flash" (http://flash.uchicago.edu/website), used for studying astrophysical thermonuclear flashes such as supernovae, is one possible test bed for this research, although part of the work will also include evaluating alternatives.
The implicit analysis part of such codes typically makes heavy use of sparse matrix linear solvers and the study of the most appropriate of these will form an important part of the research.
There are three main areas for study:
- The mis-match in parallel scalability using typical current techniques between the explicit part (scaling to thousands of processors) and the implicit part (limited to tens of processors).
- The performance and scalability of parallel sparse solvers used for the implicit solutions. This is of particular importance since the differential scalability of the explicit/implicit coding is likely to lead to the implicit solution being the performance bottleneck.
- The performance of such codes on a range of emerging computer architectures using a variety of programming techniques.
The aim is find techniques that will enable such codes to be used to solve very large 3D models and to scale to performance measured in Petaflops. This will be done using a combination of runs on existing massively parallel supercomputer facilities together with modelling and prediction techniques. However, facilities approaching Petaflops levels of performance may become available during the course of the research.
We are looking for applications from students with a good undergraduate degree in science, engineering, computer science or mathematics. The key skills required are a strong programming ability in either Fortran, C, C++ or Java, and a keen interest in solving practical problems using powerful computers. Knowledge of High Performance Computing and parallel programming would be advantageous but not essential as appropriate training is available.
The PhD will be based at the University of Edinburgh and will be co-supervised by EPCC (www.epcc.ed.ac.uk) and AWE (www.awe.co.uk). EPCC is an Institute within the School of Physics and Astronomy and the PhD qualification would be awarded by the University of Edinburgh.
Funding is available for a studentship to cover UK/EU tuition fees and a full maintenance grant for three years. Note that only British nationals are eligible for this studentship.
Applications, including a covering letter, a full C.V. and the names and contact details of two academic referees, should be submitted to jane.patterson [at] ed.ac.uk (Jane Patterson) by email.
The initial closing date for applications is Friday 15th July, with interviews planned to take place on 28th or 29th July. PhD studies will begin in September 2011.