Search NASA⌕ Search

Engineering topics

Shalf, J.

Publications and source records attributed to Shalf, J..

Report on the ASCR Workshop on Modeling and Simulation of Exascale Systems and Applications

In the last few years, numerous studies commissioned by the Department of Energy (DOE), the Defense Advanced Research Projects Agency (DARPA), and others have examined the challenges that will have to be overcome for computational science to achieve exascale, a two order-of-magnitude increase in capability beyond what is possible today. Exponential growth in the number of devices (Moore's Law) is expected to continue unabated, but the end of Dennard scaling, and hence the growth in performance of individual processors, has led to exponential growth in concurrency. Power consumption has also grown to the point where the cost of the electrical power for systems threatens to exceed the cost of acquiring them. Ever-shrinking devices, and the growth in their number, are expected to increase the rate of soft, or transient, errors such that applications will no longer be able to assume correct behavior of the underlying machine. Adapting to such systems will require new mathematical algorithms that minimize synchronization and data movement, and a new generation of scientific software. Rather than addressing these issues in isolation, a new process of "Co-Design" is being pursued in which application and computer scientists work toward the common goal of an exascale ecosystem of systems and applications. Modeling and simulation (ModSim) is a critical part of this process. It enables scientists and engineers to analyze future algorithms, applications, and computing systems, long before they are realized, and make the necessary design decisions such that exascale science is broadly achieved. Later, when exascale applications and systems are realized, ModSim technology will enable us to understand their behavior, debugging and optimizing them. Given the increasing importance of modeling and simulation, a workshop was organized by DOE/ASCR to assess the current state of the art and to identify research challenges that must be overcome in this field. Fifty experts, representing academia, industry, and government, participated in the workshop, and others contributed both beforehand and afterwards. This report summarizes their findings.

97 MATHEMATICS AND COMPUTING↗

Stencil computation optimization and auto-tuning on state-of-the-art multicore architectures.

Understanding the most efficient design and utilization of emerging multicore systems is one of the most challenging questions faced by the mainstream and scientific computing industries in several decades. Our work explores multicore stencil (nearest-neighbor) computations — a class of algorithms at the heart of many structured grid codes, including PDE solvers. We develop a number of effective optimization strategies, and build an auto-tuning environment that searches over our optimizations and their parameters to minimize runtime, while maximizing performance portability. To evaluate the effectiveness of these strategies we explore the broadest set of multicore architectures in the current HPC literature, including the Intel Clovertown, AMD Barcelona, Sun Victoria Falls, IBM QS22 PowerXCell 8i, and NVIDIA GTX280. Overall, our auto-tuning optimization methodology results in the fastest multicore stencil performance to date. Finally, we present several key insights into the architectural tradeoffs of emerging multicore designs and their implications on scientific algorithm development.

97 MATHEMATICS AND COMPUTING↗

Lattice Boltzmann simulation optimization on leading multicore platforms

We present an auto-tuning approach to optimize application performance on emerging multicore architectures. The methodology extends the idea of searchbased performance optimizations, popular in linear algebra and FFT libraries, to application-specific computational kernels. Our work applies this strategy to a lattice Boltzmann application (LBMHD) that historically has made poor use of scalar microprocessors due to its complex data structures and memory access patterns. We explore one of the broadest sets of multicore architectures in the HPC literature, including the Intel Clovertown, AMD Opteron X2, Sun Niagara2, STI Cell, as well as the single core Intel Itanium2. Rather than hand-tuning LBMHD for each system, we develop a code generator that allows us identify a highly optimized version for each platform, while amortizing the human programming effort. Results show that our autotuned LBMHD application achieves up to a 14 improvement compared with the original code. Additionally, we present detailed analysis of each optimization, which reveal surprising hardware bottlenecks and software challenges for future multicore systems and applications.

97 MATHEMATICS AND COMPUTING↗