Advances in High-Performance Computational Earth Sciences: Applications and Frameworks (IHPCES) Session 1

Time and Date: 10:35 - 12:15 on 12th June 2019

Room: 0.3

Chair: Takashi Shimokawabe

414 A Fast 3D Finite-element Solver for Large-scale Seismic Soil Liquefaction Analysis [abstract]
Abstract: The accumulation of spatial data and development of computer architectures and computational techniques raise expectations for large-scale soil liquefaction simulations using highly detailed three-dimensional (3D) soil-structure models; however, the associated large computational cost remains the major obstacle to realizing this in practice. In this study, we increased the speed of large-scale 3D soil liquefaction simulation on computers with many-core wide SIMD architectures. A previous study overcame the large computational cost by expanding a method for large-scale seismic response analysis for application in soil liquefaction analysis; however, that algorithm did not assume the heterogeneity of the soil liquefaction problem, resulting in a load imbalance among CPU cores in parallel computations and limiting performance. Here we proposed a load-balancing method suitable for soil liquefaction analysis. We developed an efficient algorithm that considers the physical characteristics of soil liquefaction phenomena in order to increase the speed of solving the target linear system. The proposed method achieved a 29-fold increase in speed over the previous study. Soil liquefaction simulations were performed using large-scale 3D models with up to 3.5 billion degrees-of-freedom on an Intel Xeon Phi (Knights Landing)-based supercomputer system (Oakforest-PACS).
Ryota Kusakabe, Kohei Fujita, Tsuyoshi Ichimura, Muneo Hori and Lalith Wijerathne
173 Performance evaluation of tsunami inundation simulation on SX-Aurora TSUBASA [abstract]
Abstract: As tsunamis may cause damage in wide area, it is difficult to immediately understand the whole damage. To quickly estimate the damages of and respond to the disaster, we have developed a real-time tsunami inundation forecast system that utilizes the vector supercomputer SX-ACE for simulating tsunami inundation phenomena. The forecast system can complete a tsunami inundation and damage forecast for the southwestern part of the Pacific coast of Japan at the level of a 30-m grid size in less than 30 minutes. The forecast system requires higher-performance supercomputers to increase resolutions and expand forecast areas. In this paper, we compare the performance of the tsunami inundation simulation on SX-Aurora TSUBASA with those on Xeon Gold and SX-ACE. SX-Aurora TSUBASA is a new vector supercomputer released in 2018 and its peak performance is 4.3 Tflop/s of single precision floating-point operations. We clarify that SX-Aurora TSUBASA achieves the highest performance among the three systems and has a high potential for increasing resolutions as well as expanding forecast areas.
Akihiro Musa, Takashi Abe, Takumi Kishitani, Takuya Inoue, Masayuki Sato, Kazuhiko Komatsu, Yoichi Murashima, Shunichi Koshimura and Hiroaki Kobayashi
315 Parallel Computing for Module-Based Computational Experiment [abstract]
Abstract: Large-scale scientific code plays an important role in scientific researches. In order to facilitate module and element evaluation in scientific applications, we introduce a unit testing framework and describe the demand for module-based experiment customization. We then develop a parallel version of the unit testing framework to handle long-term simulations with a large amount of data. Specifically, we apply message passing based parallelization and I/O behavior optimization to improve the performance of the unit testing framework and use profiling result to guide the parallel process implementation. Finally, we present a case study on litter decomposition experiment using a standalone module from a large-scale Earth System Model. This case study is also a good demonstration on the scalability, portability, and high-efficiency of the framework.
Zhuo Yao and Dali Wang
390 Heuristic Optimization with CPU-GPU Heterogeneous Wave Computing for Enhancing Three-dimensional Inner Structure [abstract]
Abstract: To increase the reliability of numerical simulations, it is important to use more reliable models. This study proposes a method to generate a finite element model that can reproduce observational data in a target domain. Our proposed method searches parameters to determine finite element models by combining simulated annealing and finite element wave propagation analyses. In the optimization, we utilize heterogeneous computer resources. The finite element solver, which is the computationally expensive portion, is computed rapidly using GPU computation. Simultaneously, we generate finite element models using CPU computation to overlap the computation time of model generation. We estimate the inner soil structure as an application example. The soil structure is reproduced from the observed time history of velocity on the ground surface using our developed optimizer.
Takuma Yamaguchi, Tsuyoshi Ichimura, Kohei Fujita, Muneo Hori and Lalith Wijerathne
383 A Generic Interface for Godunov-type Finite Volume Methods on Adaptive Triangular Meshes [abstract]
Abstract: We present and evaluate a programming interface for creating high performance Godunov-type finite volume applications with the framework sam(oa)2. This interface requires application developers only to provide problem-specific implementations of a set of operators, while sam(oa)2 transparently manages its many HPC features, such as memory-efficient adaptive mesh refinement, parallelism in distributed and shared memory and vectorization of Riemann solvers. We focus especially on the performance of vectorization, which can be either managed by the framework (with compiler auto-vectorization of the operator calls) or directly by the developers in the operator implementation (possibly using more advanced techniques). We demonstrate the interface's performance using two example applications based on variations of the shallow water equations. Our performance results show successful vectorization using both approaches, with similar performance. They also show that the applications developed with the new interface achieve performance comparable to other analogous applications developed without the new layer of abstraction, directly into the framework's core.
Chaulio R. Ferreira and Michael Bader