New international partnership to prepare for emerging simulation platforms

December 20, 2013

Argonne National Laboratory, in partnership with the public research institute for computational science Inria, the University of Illinois at Urbana-Champaign (UIUC), and the French National Center for Scientific Research (CNRS), has been named recipient of a $646,000 (including institution participation) grant from the Partner University Fund (PUF).

The three-year grant, titled “NextGN: Preparing for Next Generation Numerical Simulation Platforms,” seeks to develop new research activities and form new generations of researchers focusing on emerging exascale platforms and cloud environments.

Simulation has become a vital component in science, with many researchers relying on it for their studies. But the large number of cores, the large power consumption, and the increasing failure rates of extreme-scale platforms present major obstacles to using such simulation platforms effectively.

“We will address these obstacles by improving fault tolerance and by significantly reducing the communication costs between the computational cores, the memory, the network, and input/output,” said Franck Cappello, senior computer scientist in Argonne’s Mathematics and Computer Science Division, who will head the Argonne effort.

“Faults are a feature of the platform,” said Yves Robert, who heads the project for Inria. “One of our objectives, therefore, is to provide mathematical estimates of the expected throughput of the platform so that a user could, for example, migrate an ongoing computation to a different resource just before the fault takes place. We also plan to develop a model that will allow users to select the right combination of resilience methods for exascale.”

Another concern for high-performance computing applications is being able to create environments with predictable placement on cloud resources. In the Grid’5000 in France, for example, investigators often provision environments to study the effects of various phenomena in cloud computing. As part of the PUF project, the researchers will explore ways for controlling how virtual machines are scheduled not only on a particular cloud but also between available clouds wherever resource contention can occur.

The researchers from the four U.S. and French institutions also will investigate challenges for computing derivatives of applications aiming at exascale performance. For example, higher-order derivatives are important for uncertainty quantification applied to realistic problems. Researchers at CNRS are particularly concerned with understanding how to exploit resilience techniques for the next generation of atmospheric models. Other applications that will benefit from the research developed in the PUF project include earthquake simulation, molecular dynamics, and astrophysics.

In addition to investigating fundamental aspects such as mathematical analysis and modeling, the project includes experimental aspects – software prototype development and large-scale experimentation on the Blue Gene/Q system at Argonne, the Blue Waters system at UIUC, and the PRACE systems in Europe.  “By actively involving students and postdocs in this research program, we will form the new generation of numerical simulation experts who will improve and run applications on upcoming high-performance computing systems,” said Robert.

Indeed, a key part of the project is the education of young researchers. The PUF grant partners hope to strengthen the collaboration between Argonne, Inria and UIUC and to extend opportunities to other teams at CNRS. To this end, the grant provides significant funding for student and faculty visits in support of joint research, two research workshops each year with strong student participation, and an annual summer school opened to the students of the collaboration.

 “This is an important grant for Argonne,” said Cappello. “Only five PUF grants were awarded in this highly competitive call. And Argonne was the only U.S. national laboratory receiving this year’s award.”

For more information on the PUF grant, see the website.