Invited Talks

The following speakers already have accepted to speak at the conference. More information will be added once it is available.

European E-Infrastructure: Promoting Global Virtual Research Communities

Maria Ramalho-Natario
European Commission, INFSO

Abstract: The Framework Programme 7, through the specific Programme 'Capacities', will ensure that support to existing research infrastructures (e.g. high speed inter-networks, computing grids and digital repositories) will continue and will help to create new research infrastructures of pan-European interest. This talk will focus on the vision for FP7 of the development of ICT-based infrastructures, also named e-Infrastructure, insisting on the aims of the Calls in 2007 and early 2008 to support virtual global research communities. For the benefit of the audience, a parenthesis will be made on the plans for the creation of the European High-Performance Computing service in FP7.

Programming in the Multicore Era

Barbara Chapman
University of Houston, Texas

Abstract: Dual-core machines are now actively marketed for desktop and home computing. Systems with a larger number of cores exist, and more are planned. Some cores are capable of executing multiple threads. At the very high end, programmers need to design codes for execution by thousands of processes or threads and have begun to consider how to write programs that can scale to hundreds of thousands of threads. Clearly, the future is multi- and many-core, as well as many-threaded. In the past, most application developers could rely on Moore's Law to provide them with steady performance improvements. But we have entered an era in which they may have to expend considerable effort if their codes are to exploit the processing power offered by next-generation platforms.

Existing shared memory parallel programming APIs were not necessarily designed for general-purpose computing or with many threads in mind. Distributed memory paradigms do not necessarily allow the expression of fine-grained parallelism or provide full exploitation of architectural features. The fact that threads share some resources in multicore systems makes it hard to reason about the impact of program modifications on performance and results may be surprising. Will programmers be able to use multicore platforms effectively?

In this presentation, we discuss the challenges posed by multicore technology. We then review recent work on programming languages that are potentially interesting for multicore platforms, and discuss on-going activities to extend compiler technology in ways that may help the multicore programmer.

Simulation of Heart-Assist Devices

Marek Behr
RWTH Aachen University

Abstract: Parallel computing is enabling computational engineering analyses of unprecedented complexity to be performed. This talk reports on parallel finite element flow simulations supporting the development of implantable ventricular assist devices in the form of continuous-flow axial pumps. These pumps offer simplicity and reliability needed in long-term clinical applications. Their design however poses continuing challenges, such as high shear stress levels, flow stagnation and onset of clotting, and loss of pump efficiency.

Elevated shear stress levels are particularly evident in mechanical biomedical devices. One of the adverse responses of the red blood cells to elevated shear is hemolysis, dependent on both dose and time. The distribution of the shear stress levels in a complex flow field in a rotary blood pump chamber as well as the duration of the blood cells' exposure to these pathological conditions are largely unknown. Device designers are often compelled to make decisions about the details of pump configuration guided only by the global, time- and space-averaged, indicators of the shear stress inside the pump, such as the hemolysis observations made on the exiting blood stream. This challenge of detailed analysis and reduction of shear stress levels while maintaining pump efficiency as well as the need to pinpoint any persistent stagnation areas in the flow field motivates our current computational work.

We describe the flow simulation methodology and apply it to the problem of anylysis of blood flow in an axial ventricular assist device, the MicroMed DeBakey LVAD. This pump consists of the flow straightener, a six-bladed impeller, and a six-bladed diffuser inside a cylindrical housing. The simulations must explore a range of impeller speed and various pressure conditions. The computations are performed on an IBM Blue Gene.

Using these simulations as an illustration, we will focus on the architecture of the MPI-based finite element code, and on steps taken to ensure reasonable parallel speed-up on 4096 CPUs, including performance analysis and bottleneck identification. In view of the need for design optimization, where unsteady flow fields as well as their sensitivities with respect to the design parameters must be computed repeatedly while seeking a minimum of an flow-dependent objective function, the use of thousands of CPUs is a critical factor that makes such optimization practical.

Towards Petascale Grids as a Foundation of E-Science

Satoshi Matsuoka
Tokyo Institute of Technology

Abstract: While there is general consensus that computing platforms underlying the grid infrastructures will continue to evolve, variance in the speed of technology acceleration in HPC is causing many of the assumptions made in the early days of grid to no longer hold. Such divergence in the metrics, as well as wider proliferation of related technologies such as Web2.0, will be changing the optimal design of the overall grid infrastructure towards more centralization in the data/computing centers, as we also have experienced in the past for the Internet. Still, some facilities will remain fundamentally distributed, as simple centralization in one location might not be feasible for various reasons. Based on our recent experiences with our TSUBAME supercomputer, which is currently Asia-Pac's fastest machine according to the Top500, and its next petascale generation design thereof, we will discuss the future design of multi-petascale grids with such machines being constituent massive resource nodes instead of vast distribution.

Forschungszentrum Jülich
D-52425 Jülich