With Moore's law progressively running out of steam, heterogeneous computing architectures have been powering the top supercomputers in the world for many years and are now finding broader adoption across the industry. The trend towards sustainable computing also requires domain-specific heterogeneous hardware architectures, which promise further gains in energy efficiency. At the same time, today's high performance computing applications have evolved from monolithic simulations in a single domain to multidisciplinary complex workflows. In this paper, we explore how these trends affect system design decisions and what this means for future computing system architectures.

Computational models can define the functional dynamics of complex systems in exceptional detail. However, many modeling studies face seemingly incommensurate requirements: to gain meaningful insights into some phenomena requires models with high resolution (microscopic) detail that must nevertheless evolve over large (macroscopic) length- and time-scales. Multiscale modeling has become increasingly important to bridge this gap. Executing complex multiscale models on current petascale computers with high levels of parallelism and heterogeneous architectures is challenging. Many distinct types of resources need to be simultaneously managed, such as GPUs and CPUs, memory size and latencies, communication bottlenecks, and filesystem bandwidth. In addition, robustness to failure of compute nodes, network, and filesystems is critical.

The concept that scientists and engineers should be able to monitor and control simulations running on supercomputers has been discussed and implemented since the late 1980s. The recent explosion in the variety and capabilities of mobile devices allows this access to be taken to a new level, since the simulations running on supercomputers can be accessed as users are moving around in their normal work activities. This means that they have to be able to connect and disconnect at will from the simulation running on the supercomputer, without disturbing its execution. We present a general framework for such “mobile supercomputing” within the framework of Web services standards. To illustrate the potential of our method, we present a particular application of this framework to provide a rich mobile interface to a lattice-Boltzmann simulation of complex fluids running on an IBM Blue Gene/Q supercomputer.

SUMMARY Computational fluid dynamics is an increasingly important application domain for computational scientists. In this paper, we propose and analyze optimizations necessary to run CFD simulations consisting of multibillion‐cell mesh models on large processor systems. Our investigation leverages the general industrial Navier–Stokes CFD application, Code_Saturne, developed by Electricité de France for incompressible and nearly compressible flows. In this paper, we outline the main bottlenecks and challenges for massively parallel systems and emerging processor features such as many‐core, transactional memory, and thread level speculation. We also present an approach based on an octree search algorithm to facilitate the joining of mesh parts and to build complex larger unstructured meshes of several billion grid cells. We describe two parallel strategies of an algebraic multigrid solver and we detail how to introduce new levels of parallelism based on compiler directives with OpenMP, transactional memory and thread level speculation, for finite volume cell‐centered formulation and face‐based loops. A renumbering scheme for mesh faces is proposed to enhance thread‐level parallelism. Copyright © 2012 John Wiley & Sons, Ltd.

Computational Fluid Dynamics (CFD) is an increasingly important application domain for computational scientists. In this paper, we propose and analyze optimizations necessary to run CFD simulations consisting of multi-billion-cell mesh models on large processor systems. Our investigation leverages the general industrial Navier-Stokes open-source CFD application, Code_Saturne, developed by Electricité de France (EDF). Our work considers emerging processor features such as many-core, Symmetric Multi-threading (SMT), Single Instruction Multiple Data (SIMD), Transactional Memory, and Thread Level Speculation. Initially, we have targeted per-node performance improvements by reconstructing the code and data layouts to optimally use multiple threads. We present a general loop transformation that will enable the compiler to generate OpenMP threads effectively with minimal impact to overall code structure. A renumbering scheme for mesh faces is proposed to enhance thread-level parallelism and generally improve data locality. Performance results on IBM Blue Gene/P supercomputer and Intel Xeon Westmere cluster are included.

The multi-core Cell BE architecture has the potential to deliver high performance computation to many applications. However, it requires parallel programming on several levels and uses a memory model that requires explicit scheduling of memory transfers. The RapidMind development platform enables access to the power of the Cell BE by providing a simple data-parallel model of execution. It combines a dynamic compiler with a runtime streaming execution manager, and provides a single system image for computations running on multiple cores. The interface to this system is embedded inside ISO standard C++, where it can capture computations specified through an innovative metaprogrammed interface. This tutorial will introduce and demonstrate the use of the RapidMind platform on the Cell BE through a number of examples drawn from visualization and scientific computation.