UM2 and HPC@LR create a hybrid supercomputer

To read the French version of the case study click here.

Montpellier 2 University (UM2), one of the leading educational institutions in France, traces its history back to 1808. Today, UM2 has more than 15,000 students and 3,000 tenured teaching staff. The institution offers both undergraduate and postgraduate/research degree courses across all major scientific and technological fields.

UM2 is one of the members of the public-private consortium that runs the Languedoc-Roussillon high-performance computing center (known as HPC@LR). HPC@LR brings together supercomputing resources, technical expertise and educational facilities to boost the capacity for scientific analysis—with a particular focus on the environment, water and health. Its resources are available not only to university researchers, but also to local companies.

Anne Laurent, Associate Professor, Laboratory of Informatics, Robotics and Microelectronics at UM2, comments: “Our belief is that massively parallel analysis will be the leading approach in scientific research in the future. We wanted a hybrid supercomputing architecture that would give our researchers, students and business users experience of the latest technologies, and IBM was the only vendor able to give us everything we needed. Our solution includes three different CPU architectures—Intel® Xeon®, IBM Cell Broadband Engine and IBM POWER7® — as well as GPU cores, but the hybrid concept of the solution means that they are accessible as a single unit. The cluster storage management software, IBM General Parallel File System, is a key element of this unified access.”

All-in-one solution
The hybrid IBM supercomputer managed by HPC@LR is hosted at CINES, on the Saint-Priest campus of UM2. Based on the IBM BladeCenter and IBM System x iDataPlex platforms, it uses low-latency, high-bandwidth InfiniBand networking technology to move data rapidly between the processors and 150 TB of storage. The 84 iDataPlex dx360 M3 nodes each contain two six-core Intel Xeon Processor 5600 series CPUs for a total of 1,008 cores of processing power. This environment also includes 12 GPU (graphics processing unit) cards, which slot into the iDataPlex machines to handle massively parallel computations. The second part of the supercomputer is based on IBM BladeCenter, with four BladeCenter QS22 blades and one PS702 Express blade. The QS22 blades each feature two IBM PowerXCell™ 8i processors with one PPE core and eight enhanced double precision SPE cores, while the PS702 features four quad-core IBM POWER7 processors.

“What we wanted was a completely seamless solution combining hardware and software,” says Anne Laurent. “This massively multicore IBM solution delivers precisely that, acting as a coherent set of resources from the computational point of view, thanks in particular to an advanced system for managing batches of jobs submitted to the supercomputer.”

The use of IBM General Parallel File System (GPFS™) also helps to unite the computing resources into a coherent whole, ensuring that each user can access all data across all the different elements of the supercomputer. IBM Business Partner SERVIWARE, an HPC specialist, took part in the planning and installation of the solution.

Anne Laurent comments: “The advanced nature of the solution means that not everyone can use it without some additional training and assistance. Here at HPC@LR, we draw on the complementary skills of IBM and CINES to help researchers and companies develop their skills around HPC. In particular, this means helping them to code and run their programs not only on the local infrastructure but also on national and European resources. For the GPU element of the solution, we work with specialist partners such as ASA and HPC Project. All of this assistance enables researchers, universities and companies to optimize the parallelization of their programs as they port their analytical tools to the computing architectures of tomorrow.”

Taking on real-world challenges
A number of different research teams and businesses are using the resources at HPC@LR, taking advantage of the massive computing power to process large volumes of complex data at high speed. One such research organization is OREME (the Mediterranean Environmental Research Observatory). The observatory focuses on the tracking of the effect of natural hazards, global change and anthropic side effects on Mediterranean media. OREME rests on a dense network of multiparameter stations of continuous observation of physical, chemical and biological status of natural media, and most notably: the geophysical monitoring of ground and surface waters, GPS monitoring of ground movements and tropospheric humidity, the monitoring of reference animal and vegetal species and overall biological state of selected areas, and the monitoring of coastal environment. Using resources such as data warehouse, cloud strategies and HPC@LR, OREME allows complex analysis of these large environmental dataflows revealing correlations between conditions and events that otherwise remain hidden.

Another user is BRL, a regional group specializing in irrigation, which employs a rich set of simulation models to design irrigation and drainage projects. The extreme computational power of HPC@LR has enabled BRL to open up a new market around real-time modeling for water pollution events. For example, a 30-hour water pollution event that would previously have required six hours of computing time now requires just three minutes. This means that the models created by BRL can actually guide operational activities, rather than just providing analysis after the event.

In other areas, too, HPC@LR is making a significant contribution to attempts to model and understand local physical phenomena, including the effects of floods caused by the short, intense bursts of rainfall that characterize the Languedoc region. The expectation is that urban planners and city authorities will be able to understand the likely impact of flooding and take the appropriate measures to protect neighborhoods.

The recent establishment of a center of water excellence at IBM Montpellier has provided HPC@LR with additional resources for HPC analysis, and the institution also shares its research with the Barcelona Supercomputing Center (BSC).

“HPC skills are comparatively rare within France, and it is even harder to find them in close proximity,” comments Anne Laurent. “The IBM HPC center of excellence at IBM Montpellier is an enormous benefit—not only for our own research and for local companies, but also because of the link that it gives us to IBM Research and to the BSC in particular. The data sets we are attempting to analyze are increasingly voluminous, and we are seeking to understand them at ever finer levels of detail. Because of this, the corresponding complex systems that we need to model and simulate require massively parallel architectures, as well as the skills to port code to them. IBM has helped us to meet both requirements, with a powerful hybrid supercomputer and deep coding expertise.”