Petascale computing, which is at the heart of the PRACE project, was one of the main topics addressed in workshop. One petaflop/s means computer performance of 10 to the power of 15 flop/s. The current status of the PRACE project, including the progress that has been made in terms of hardware infrastructure, software applications and training, was discussed during the workshop. Mauricio Araya and Ricardo Fonseca were amongst those who presented their own research.

Research in plasma physics and seismic imaging…

Professor Ricardo Fonseca, researcher at the Instituto Superior Técnico in Lisbon, Portugal, works in the field of Plasma physics, focusing in particular on high energy density plasmas and their various applications.

“Plasma-based technology has been at the forefront of scientific research for more than half a century, with applications in a large number of different fields, far beyond the domain of fundamental science, ranging from engineering to medicine,” Fonseca explains.

“The work we are carrying out at the Instituto Superio Técnico covers many aspects of plasma physics research, from laser plasma interaction to astrophysics. In particular, the development of compact plasma-based accelerators is of great interest to us. It has applications in, for example, fundamental particle physics research, table-top narrowband x-ray sources, and free-electron lasers; along with many others in medicine (e.g. cancer therapy), biology and the material sciences. Moreover, we conduct research in astrophysical plasmas and space weather, with applications ranging from magnetospheric physics to spacecraft design (e.g. improving the robustness of spacecrafts against solar wind); and we also work in inertial confinement nuclear fusion, in which high-power lasers confine and ignite the fusion plasma, with the ultimate goal of moving towards a new energy source free from oil dependency,” Fonseca continues.

Dr. Mauricio Araya is a researcher in computational geophysics at the Barcelona Supercomputing Center, Spain, and works on seismic imaging, a computational process that generates images of the Earth’s subsurface structure.

“Our research is focused on a particular set of non-invasive seismic techniques, in which a set of acoustic data from recorded wave fields is capable of providing a model of the subsurface. Basically, we simulate the propagation of acoustic waves through the Earth’s subsurface, and through the reflection and refraction of these waves, we are able to create images of the inner structures of the Earth. Given that the simulations are extremely time-consuming, the main goal of our research is to speed up the seismic imaging process and, at the same time, to increase the accuracy of the resulting images. The techniques that we are developing are of particular interest for both the geophysical community and the oil-industry: for example, they assist the latter in making decisions regarding where to drill. Bearing in mind that each drilling operation costs between 100 and 150 million dollars, they are of huge importance to this industry and, indeed, to the energy sector more generally, as oil is one of the most important sources of energy today,” Araya explains.

Collision of two relativistic plasma slabs. Professor Ricardo Fonseca works in the field of Plasma physics, focusing in particular on high energy density plasmas and their various applications. © GoLP/IPFN/Instituto Superior Técnico.

…is heavily dependent on computational resources

What brings together the projects of Fonseca and Araya, despite the different natures of their respective area of interest, is the fact that both require huge amounts of computational resources.

Fonseca’s research on plasmas, for example, relies heavily upon numerical simulations.

“The phenomena that we are interested in are extremely complex, and the full dynamics of such systems can only be investigated through the use of numerical simulations,” says Fonseca. “These ‘numerical experiments’ cannot be run on standard computers; they require advanced computing power and are conducted on supercomputers. The process mimics that of experimental physics: a ‘numerical experiment’ is prepared and submitted to a supercomputer on which it is run; we then obtain the results, analyze them, and decide on the next experiment. Many important advances have been made in this virtual laboratory, specifically in the field of laser-plasma accelerators, focusing on the production of high quality electron beams. Working on the basis of these outcomes, experimental teams have conducted laboratory experiments and confirmed our numerical results, demonstrating the feasibility of compact, tabletop accelerators that use high intensity lasers for applications in medicine, imaging technology and high energy physics,” Fonseca continues.

Huge computational resources are obviously also required in Araya’s project, given that its main aim is to speed up the seismic imaging process.

“Complex simulations of geophysical systems for huge data sets require large High Performance Computing (HPC) capability,” Araya explains.

“Only supercomputers provide the amount of computational power required by the task at hand. In our case, at the Barcelona Supercomputing Center, we are already working on two very powerful supercomputers: a terascale supercomputer (MareNostrum) and a PRACE prototype supercomputer (MariCel). Thanks to the combination of efficiently developed algorithms and access to these HPC facilities we are currently able to speed up the computing of the simulations by one order of magnitude – meaning that the execution time of our typical simulations has been reduced to 10% of what it used to be,” he continues.

Plasma physics and Seismic imaging in the era of Petascale computing

The two researchers see the development of computational capacities as a crucial tool for their respective fields. It is not, then, surprising to find that they are heavily involved in the PRACE project. Both are convinced that petascale computers will offer important new opportunities for their own areas of research.

“Petascale computing opens a myriad of possibilities, not only because it will reduce execution times, but also because such massive amounts of computational power will allow us to push the algorithms and mathematical schemes to a higher level of accuracy,” says Araya.

“Two other benefits can be expected: firstly, we will be able to conduct simulations in a time-frame of days rather than months, meaning that important decisions on the results can be taken much more quickly; secondly, while current simulations tend to simplify aspects of reality due to computational limitations, with petascale computing some of these simplifications can be discarded, enabling us to obtain more accurate and realistic results,” Araya explains.

Fonseca also emphasizes the new opportunities that petascale computers will bring in plasma physics.

“The simulation models that we use are very detailed, and rely on solving the microscopic physics involved. This means that our models are very accurate but that they are also very computationally expensive. One-to-one modeling of large experiments or scenarios, in which one needs to model simultaneously both very small and very large spatial/ temporal scales, therefore requires petascale computing. Specifically, such simulations need to be run for tens of millions of timesteps; they require over a terabyte of computer memory; and they produce tens of terabytes of results. The PRACE project represents an excellent opportunity for us to gain access to the required computational resources in Europe,” Fonseca concludes.

Petascale computing also gives new challenges

Both researchers, however, also emphasize the new challenges that will have to be faced in moving to petascale computing. While Araya points to the fact that the different codes will have to scale well to larger environments and that many important algorithms will have to be enhanced, Fonseca notes that petascale computing is at a higher level than normal High Performance Computing, thus raising the bar for all applications that aim to use it. These challenges will have to be met in order to fully exploit the outstanding computational resources offered by petascale computers.

Damien Lecarpentier