To speak of supercomputing is to speak of the oil industry, air quality prediction and the design of the computer architecture of the future. More and more scientific and technological projects are requiring the use of intensive calculation.
Supercomputers have become an indispensable tool for scientific progress. The great computing power of infrastructures such as MareNostrum, which allow the processing of millions of instructions per second and the simulation of highly complex phenomena that cannot be reproduced in a laboratory, is fundamental for providing answers in different fields of science and technology.
MareNostrum, one of the most powerful supercomputers in Europe, which has a processing capacity of 94.21 teraflops (94.21 trillion operations per second), is located at the Barcelona Supercomputing Center-Centro Nacional de Supercomputación (BSC-CNS) directed by Mateo Valero, holder of a PhD in Telecommunications Engineering and professor at the UPC. Since it was launched in 2005 it has provided support to numerous high-quality research projects of international impact.
One of the initiatives the BSC-CNS is currently working on is the Kaleidoscope project, developed in collaboration with Repsol. The project works to generate seismic images of the subsoil, which are then processed 14 times faster than with other technologies in order to determine the viability of offshore oil extraction in the Gulf of Mexico.
The oil industry uses algorithms to analyze and interpret the maps generated by seismic imaging. Technically, what differentiates this project from others is that the team, led by José M. Cela, member of the BSC-CNS and professor at the UPC, uses the RTM (reverse time migration) algorithm, currently the method that generates the most accurate images of the physical world. "We have created the first commercial RTM", explains Cela. "Basically, what we've managed to do is accelerate the algorithm, which can now be run in a few hours, thus creating a functional technique that Repsol can use for production. And we've done so using all possible levels of parallelism", he explains.
The first version of the RTM was developed using the MareNostrum JS21 processors, 4,000 of which were working on it for three months. The following step was to migrate the algorithm to another platform, the Cell processor. The project code was executed using MariCel-a supercomputer prototype based on Cell and developed by the BSC-and now the algorithm can be computed in just a few hours.
This technology puts Repsol at the vanguard in exploration of geologically complex areas where the presence of saline layers act as mirrors that prevent a view of what is underneath if using conventional technology. Proof of the impact the project has had is the multinational's great exploratory success in 2009.
Predicting air quality
The project, dubbed Caliope, is funded by the Spanish Ministry of the Environment and coordinated by José María Baldasano, director of the BSC's Earth Sciences Section and professor at the UPC. It aims to develop a system to predict air quality in Spain.
To develop a modeling system like this one, three key elements have been taken into account: meteorology, emissions of pollutants and atmospheric dispersion and reaction. This great amount of data is processed thanks to the computational capacity of MareNostrum, until images are obtained that show a forecast of air quality for the next 48 hours.
To add the necessary improvements to the model developed, the results obtained are compared with the real data obtained through the different measuring stations.
Another aspect specific to this project and which makes Caliope different is the fact that it is designed to work with a 12-kilometer grid resolution over Europe and a four-km one over the Iberian Peninsula and the Balearic Islands. This level of detail places it at the vanguard of such systems, since the majority currently work with grids of 10-20 km2.
Moreover, in contrast to other systems, Caliope involves a great many elements. "It is a highly complex project", explains Baldasano, "which employs atmospheric science and chemistry to keep the models adjusted; elements of supercomputing to make them function faster and provide daily forecasts; a great deal of management, analysis and interpretation of data to obtain a detailed emissions inventory; and finally, a visualization, communication and data management effort is also required to verify that everything is operating properly. All of this is done because we not only model data, but we also constantly compare the results provided by the model with real observations".
The end users of this pioneering system accessible via internet are the Public Administrations, responsible for managing air quality, the population at large and the world of research and development.
The aim: to gain a powerful information and management tool. After three years running, Caliope will be fully operative by July.
To enable European researchers to meet major scientific and technological challenges using supercomputing is the strategic mission of the PRACE (Partnership for Advanced Computing in Europe) consortium, which needs to consolidate Barcelona as the epicenter of supercomputing on the continent, with the BSC-CNS as the Partnership's main actor.
The PRACE will create a permanent office for high-performance supercomputing for Europe that will lead to the installation of computers with a computing power much higher than that of current supercomputers. The office will coordinate three to five centers, among them the BSC-CNS, supported by local and national supercomputing institutions that will work jointly.
Dangerous incursions of Saharan sand
Sandstorms or natural dust are a habitual phenomenon in arid zones of our planet, especially in the Sahara and Gobi deserts and those of Australia, which have particularly negative effects on atmospheric pollution or health, as they transport particles that aggravate chronic respiratory diseases.
The Mediterranean area suffers regular episodes of Saharan dust intrusions, which is why the Spanish Meteorological Agency (AEMET) has a storm prediction system to ascertain in advance its characteristics and the areas to be affected in order to take the necessary precautionary measures.
The operational system that models the transport of natural dust from the Sahara to Europe was developed by BSC-CNS (photo 2). Its research has led the World Meteorological Organization to promote a global framework for storm prediction and alert in order to obtain comparative forecasts from different models. The system is configured based on the observations provided by NASA satellites and a network of ground-based remote sensing stations, data which is processed with the help of supercomputers like MareNostrum.
MareIncognito: the vertiginous speed of supercomputing
The BSC-CNS is also working on the MareIncognito project, a research project in conjunction with IBM to define the characteristics and design of the new generation of supercomputers, which will reach a computing capacity of 10 petaflops (10,000 trillion operations per second), that is, one hundred times greater than MareNostrum's current capacity.
The head of the project is Jesús Labarta, of the UPC's Department of Computer Architecture and director of the Science and Computation Department at the BSC-CNS. "The aim", explains Labarta, "was to study the possibilities of designing machines based on the Cell processor developed by IBM in conjunction with Sony and Toshiba to equip the Play Station 3 video console. Aware of the difficulties that this involved, you could say we made a virtue of necessity: in attempting to make things work with Cell, we've reached more interesting solutions developing technology for more general applications. At the same time, we are collaborating with the team that's designing the next version of the Cell processor to see what characteristics it should have".
This is not the only line of work in a project of the scope of MareIncognito. Another team is studying programming models, that is, how these machines should be programmed in the future. "The aim is to be sure of the important concepts. The programmers should abstractly define what needs to be computed, such that the software, that is, how this is translated or is mapped on a specific hardware, is as automatic as possible. Conventional programs do not usually offer these possibilities, and this makes them difficult to migrate to new architectures", he indicates.
After a year of developing the performance analysis and simulation infrastructures, followed by another year to complete it and start obtaining results, we now need to apply it to real situations, to real programs.