Program

• Day 1 (Wednesday, January 12th)
• Day 2 (Thursday, January 13th)
• Day 3 (Friday, January 14th)
• Poster Session (Day 1)

Day 1 (Wednesday, January 12th)

All times are in JST.

9:00-9:15 Opening (Day1)

• Session Chair: Miwako Tsuji

9:15-9:20 Special Session 1

• Session Chair: Miwako Tsuji

9:15-9:20 "Memory of Hiroshi Nakashima and Putchong Uthayopas"

• Speaker: Taisuke Boku

9:20-10:10 Keynote 1

• Session Chairs: Kengo Nakajima (primary), Daisuke Takahashi (secondary)

9:20-10:10 "Scalable Solvers Software for Exascale Computers"

• Speaker: Ulrike Meier Yang (Lawrence Livermore National Laboratory (LLNL))
• Abstract (click to expand):
  The development of exascale computer architectures provides application developers the opportunity to perform larger simulations achieving more accurate solutions than ever before. To make the best use of these capabilities, application developers will need to rely on a variety of high-quality mathematical libraries that are developed by diverse independent teams throughout the HPC community. The extreme-scale scientific software kit (xSDK) project in the Exascale Computing Project (ECP) includes a variety of math libraries, whose software quality is ensured through the compatibility of the libraries with the xSDK community policies which ensure software quality as well as the seamless combined build of these libraries. The kit includes several libraries that are focused on solving large sparse linear systems, which are required by many applications. Well-designed algebraic multigrid methods, specifically when combined with Krylov methods, can provide excellent scalability and performance. However, their implementation on these new architectures provides many challenges. This talk will briefly describe the xSDK, then focus on some xSDK libraries that provide scalable linear solvers, specifically algebraic multigrid methods. We will discuss portability strategies, implementation challenges and solutions developed to achieve good performance on heterogeneous architectures for such methods and demonstrate their use within several application codes.
• Bio (click to expand):
  Dr. Ulrike Meier Yang is the Mathematical Algorithms & Computing group lead in the Center for Applied Scientific Computing at Lawrence Livermore National Laboratory. She leads the xSDK (extreme-scale scientific software kit) project in the US Department of Energy’s Exascale Computing Project and the Solution of Linear and Nonlinear Systems of Equations area of the SciDAC-5 FASTMath Institute. Her research interests are numerical algorithms, particularly iterative linear system solvers and algebraic multigrid methods, parallel computing, performance evaluation and scientific software design. She has contributed significantly to the hypre software library, which received an R&D 100 award in 2007. She is currently an associate editor for the SIAM Journal on Matrix Analysis and Applications and the Journal of Parallel and Distributed Computing. Prior to joining LLNL in 1998, she worked in the Center for Supercomputing Research and Development at the University of Illinois at Urbana-Champaign (1985-1995) and in the Central Institute of Applied Mathematics at the Research Centre Jülich, Germany (1983-1985). She earned her Ph.D. in Computer Science at the University of Illinois at Urbana-Champaign in 1995 and a Diploma in mathematics at the Ruhr-University Bochum, Germany, in 1983.
• iCalendar

10:10-10:25 Morning Break

10:25-11:55 Best Paper Finalists

• Session Chairs: Osni Marques (primary), Kento Sato (secondary)

10:25-10:55 "Performance Evaluation of Lattice Boltzmann Method for Fluid Simulation on A64FX Processor and Supercomputer Fugaku"

• Authors: Seiya Watanabe and Changhong Hu (Kyushu University)
• Abstract (click to expand):
  The lattice Boltzmann method has recently become popular as an alternative to Navier-Stokes solvers for large-scale fluid simulations. We conduct a performance study of the lattice Boltzmann method on the A64FX Arm-based processor of the supercomputer Fugaku. We compared four types of data layouts: SoA, AoS, Clusterd SoA (CSoA), and CSoA2, and three algorithms for the streaming step: Pull, Push, and Swap schemes. The performance measurement on a single CMG (Core Memory Group) shows that the combination of the CSoA2 layout and the Swap scheme has the highest performance of 176 GFLOP, which corresponds to 11.5% of the single-precision peak performance of the A64FX. Our simulations have demonstrated good weak scaling up to 16,384 nodes and achieved high performance 10.9 PFLOPS in single precision. The strong scalability is also a good result, with parallel efficiencies of 63.9%, 68.3% and 72.7% for the D3Q15, D3Q19 and D3Q27 velocity model, respectively when scaling from 512 to 16,384 nodes.
• Paper (ACM Digital Library)

10:55-11:25 "High Performance Gravitational N-body Simulations on Supercomputer Fugaku"

• Authors: Tomoaki Ishiyama (Chiba University), Kohji Yoshikawa (University of Tsukuba), Ataru Tanikawa (The University of Tokyo)
• Abstract (click to expand):
  We report performance results of large astrophysical N-body simulations on supercomputer Fugaku. We use the hybrid TreePM algorithm, in which the short-range force is evaluated by the oct-tree method, and the long-range force is solved by the particle-mesh method. We introduce three innovations that significantly improve the performance on the novel CPU and extremely large system: the near-ultimate optimization of gravity kernel for short-range forces, the domain decomposition with accelerated communications in hardware-oriented ways, and a novel communication algorithm for long-range forces. We demonstrate that our code scales near linearly up to 158,976 nodes (7,630,848 CPU cores) of Fugaku and enables us to integrate about 1.45 trillion particles per second, which is the world fastest time-to-solution. The average performance achieved is 95.82 Pflops, which corresponds to ~10% of the peak in single precision.
• Paper (ACM Digital Library)

11:25-11:55 "152K-computer-node parallel scalable implicit solver for dynamic nonlinear earthquake simulation"

• [Best Paper Award]
• Authors: Tsuyoshi Ichimura and Kohei Fujita (The University of Tokyo, RIKEN); Kentaro Koyama (Fujitsu Ltd.); Ryota Kusakabe and Yuma Kikuchi (The University of Tokyo); Takane Hori and Muneo Hori (Japan Agency for Marine-Earth Science and Technology); Lalith Maddegedara (The University of Tokyo); Noriyuki Ohi, Tatsuo Nishiki, and Hikaru Inoue (Fujitsu Ltd.); and Kazuo Minami, Seiya Nishizawa, Miwako Tsuji, and Naonori Ueda (RIKEN)
• Abstract (click to expand):
  We have used data learning and low-precision computation to develop an implicit solver that demonstrates high performance up to 152,352 computer nodes (609,408 MPI processes x 12 OpenMP threads = 7,312,896 parallel computation) and conducted an unprecedented ultra-large-scale analysis of ultra-high-fidelity fault-structure systems using nonlinear dynamic finite element analysis on three-dimensional low-order unstructured elements. The developed solver achieved 25.45-fold speedup from the state of the art solver on Fugaku and attained weak scaling efficiency of 93.7% from 9.391 billion DOF@578 computer nodes to 1.201 trillion DOF@73,984 computer nodes on performance measurement problems. Moreover, a realistic 324 billion DOF application example, which is difficult to obtain performance for, was computed in high performance. Since the developed solver is based on a highly generalizable algorithm, it is expected to contribute not only to earthquake simulation on Fugaku but also to the enhancement of similar applications in other fields and on other supercomputers.
• Paper (ACM Digital Library)

11:55-13:25 Lunch Break

13:25-14:40 Poster / Vendor Blitz 1

• Session Chairs: Takeshi Fukaya, Takaaki Miyajima
• After vendor talks (13:25-13:55), posters #102 to #114 (in ascending order) (poster program)

14:40-14:55 Afternoon Break

14:55-16:10 Poster / Vendor Blitz 2

• Session Chairs: Takeshi Fukaya, Takaaki Miyajima
• Posters #115 to #135 excepting #119 and #126 (in ascending order), then #119 and #126 (poster program)

16:10-16:15 Closing (Day 1) & Opening for Poster Session & Vendor Exhibition

• Kento Sato; Osni Marques; Takeshi Fukaya; Shinji Sumimoto; Takaaki Miyajima

16:15-17:45 Poster Session & Vendor Exhibition

• Session Chair: Takeshi Fukaya
• Poster program
• Platinum Sponsors: DataDirect Networks, Inc. (DDN), Intel, Fujitsu Limited
• Gold Sponsor: Oracle Corporation Japan
• Silver Sponsors: Advanced Micro Devices, Inc. (AMD), Altair Engineering, Inc., NVIDIA Japan
• Bronze Sponsors: NEC Corporation, Pacific Teck Japan

Day 2 (Thursday, January 13th)

All times are in JST.

9:00-9:15 Opening (Day2)

• Kento Sato; Osni Marques

9:15-10:05 Keynote 2

• Session Chairs: Rio Yokota (primary), Takeshi Iwashita (secondary)

9:15-10:05 "Challenges to large-scale weather/climate simulation for disaster prevention and mitigation"

• Speaker: Hisashi Yashiro (National Institute for Environmental Studies (NIES))
• Abstract (click to expand):
  Prevention and mitigation of meteorological disasters are important for protecting the lives and property of the people. Constant worldwide observations and numerical simulations support today's weather forecasts, and data assimilation technology combines them to improve forecast accuracy. Atmospheric simulation is a complex of various physical process modeling far from first-principles calculations. We want to improve simulation performance over a broad target range of spatiotemporal scales, from regional weather forecasts to global climate change predictions. It is always necessary to develop software in collaboration with the technological innovation of HPC. In this presentation, we will introduce our research activities in the MEXT's "Program for Promoting Researches on the Supercomputer Fugaku" (Large Ensemble Atmospheric and Environmental Prediction for Disaster Prevention and Mitigation), along with the grand challenge experiment, which was recognized as a finalist for the ACM Gordon Bell Prize 2020. In addition, we will introduce research on the fusion of atmospheric simulation and machine learning, which is one of the integration studies of (Simulation + Data + Learning).
• Bio (click to expand):
  Dr. Hisashi Yashiro is a senior researcher at the National Institute for Environmental Studies (NIES) in Tsukuba, Japan. Currently, he is a member of the project of Greenhouse gases Observing Satellite (GOSAT). He obtained a Ph.D. degree in geophysics (atmospheric science) from Tohoku University. From 2011 to 2019, he was a member of RIKEN R-CCS as a research scientist, worked on weather/climate modeling, and contributed to the large-scale weather simulation on the K computer. He is a core developer of the Nonhydrostatic ICosahedral Atmospheric Model (NICAM), a Japanese exascale climate model. He worked on the system-application co-design between the computational science of the weather/climate domain and the supercomputer Fugaku as a member of the FLAGSHIP2020 project in RIKEN R-CCS.
• iCalendar

10:05-10:20 Morning Break

10:20-11:50 Algorithms for Simulations

• Session Chairs: Dongwook Lee (primary), Osni Marques (secondary)

10:20-10:50 "Low/Adaptive Precision Computation in Preconditioned Iterative Solvers for Ill-Conditioned Problems"

• Authors: Masatoshi Kawai (The University of Tokyo) and Kengo Nakajima (The University of Tokyo, RIKEN R-CCS)
• Abstract (click to expand):
  Double precision(FP64) is generally used in computer science, and an incomplete Cholesky preconditioned conjugate gradient(ICCG) method, which is widely used in computer simulations, is a typical example. Recently, the use of single-precision(FP32) in the ICCG method with well-conditioned problems is discussed for reducing computational time and power consumption. Also, the use of half-precision(FP16) is examined in the field of machine learning, and hardware support of FP16 on GPUs and some CPUs is advancing. Using FP16 can further reduce the amount of memory translation from FP32, so high effects can be expected if it can be used with the ICCG method. When we apply FP16 to the ICCG method, It is difficult to solve the ill-condition problem because of the poor expressiveness of the exponent of FP16. Therefore, in this study, we evaluate the usefulness of FP16 in the ICCG method with several conditions of the problem, and we also proposed the implementation for CPU of and evaluated FP21 and FP42 adaptive precision whose expressiveness is higher than FP16. Additionally, to reduce the calculation time by using low-precision, it is necessary to maintain high access to the memory by effective SIMDization. Therefore, in this study, we also evaluate the performance of ELLPACK(ELL) and Sell-C-sigma storage formats, which can efficiently use SIMDization.
• Paper (ACM Digital Library)

10:50-11:20 "Assignment of idle processors to spatial redistributed domains on coarse levels in multigrid reduction in time"

• Authors: Ryo Yoda (The University of Tokyo), Matthias Bolten (University of Wuppertal), Kengo Nakajima (The University of Tokyo), and Akihiro Fujii (Kogakuin University)
• Abstract (click to expand):
  Parallel-in-time approaches for time-dependent PDEs have attracted a great deal of attention in the context of massively parallel environments. One such approach, multigrid reduction in time (MGRIT), extracts temporal parallelism by applying the reduction-based multigrid method for the temporal domain. To make efficient use of this parallelism, some processors need to be idle on a temporal coarse level of MGRIT. This paper proposes an optimization of the implementation that uses idle processors by assigning the spatial redistributed domain on coarse levels. It accelerates coarse-level spatial solvers and promotes early switching to parallelization-in-time by reducing its overhead. Although the spatial redistribution is contrary to the motivation of parallel-in-time approaches, it is expected to be effective when we need to balance spatial and temporal parallelism. This is because parallelization-in-time is already usually switched on when parallelization-in-space starts to saturate, so one can still benefit from additional parallelism. Numerical experiments demonstrate an improvement in runtime, at most, about 16.6% compared with pure MGRIT, which assigns best spatial and temporal parallelism at specific parallelism.
• Paper (ACM Digital Library)

11:20-11:50 "Massively Parallel Molecular-Continuum Flow Simulation with Error Control and Dynamic Ensemble Handling"

• Authors: Vahid Jafari, Niklas Wittmer, and Philipp Neumann (Helmut Schmidt Universität/Chair for High Performance Computing)
• Abstract (click to expand):
  In coupled molecular-continuum flow simulations, molecular dynamics (MD) simulations exhibit thermal fluctuations. Finding a way to minimize the impact of these fluctuations on the CFD solver, e.g. in terms of stability, and to control the corresponding statistical error plays a key role in order to obtain reliable results.
  In this paper, statistical error analysis is employed for MD simulations to determine the statistical error in flow velocities and the number of MD data samples to bound this error. The corresponding error estimator is augmented by a dynamic ensemble handling approach, which allows to couple a variable number of MD simulation instances to a single CFD solver. The ensemble members can be simulated independently from each other over separate coupling time intervals, enabling a high level of (MPI-based) parallelism. Adding or removing MD simulations to/from the ensemble allows to regulate the error and keep it under a prescribed threshold. All functionality is implemented in the massively parallel macro-micro-coupling tool (MaMiCo). We validate our approach by coupled molecular-continuum Couette flow simulation for liquid argon and provide scalability tests on up to 131.072 cores. The computational overhead for handling the dynamic MD ensemble is found to be rather negligible.
• Paper (ACM Digital Library)

11:50-13:20 Lunch Break

13:20-14:20 System Software for Machine Learning

• Session Chairs: Jerry Chou (primary), Yuan He (secondary)

13:20-13:50 "A Deep Neural Networks ensemble workflow from hyperparameter search to inference leveraging GPU clusters"

• Authors: Pierrick Pochelu (TotalEnergies SE); Serge G. Petiton (Univ. Lille, CNRS, UMR 9189 CRIStAL); and Bruno Conche (TotalEnergies SE)
• Abstract (click to expand):
  Automated Machine Learning with ensembling (or AutoML with ensembling) seeks to automatically build ensembles of Deep Neural Networks (DNNs) to achieve qualitative predictions. Ensemble of DNNs are well known to avoid over-fitting but they are memory and time consuming approaches. Therefore, an ideal AutoML would produce in one single run time different ensembles regarding accuracy and inference speed.
  While previous works on AutoML focus to search the best model to maximize its generalization ability, we rather propose a new AutoML to build the larger library of accurate and diverse individual models to then construct ensembles. First, our extensive benchmarks show asynchronous Hyperband is an efficient and robust way to build a large number of diverse models to combine them. Then, a new ensemble selection method based on multi-objective greedy algorithm is proposed to generate accurate ensembles by controlling their computing cost. Finally, we propose a novel algorithm to optimize the inference of the DNNs ensemble in a GPU cluster based on allocation optimization. The produced AutoML with ensemble method shows robust results on two datasets using efficiently GPU clusters during both the training phase and the inference phase.
• Paper (ACM Digital Library)

13:50-14:20 "Development of a coupler h3-Open-UTIL/MP"

• Authors: Takashi Arakawa (Research Organization for Information Science and Technology); Hisashi Yashiro (National Institute for Environmental Studies); and Kengo Nakajima (Information Technology Center, the University of Tokyo)
• Abstract (click to expand):
  We are developing a coupler, h3-Open-UTIL/MP, as part of the h3-Open-BDEC project. The h3-Open-UTIL/MP has an ensemble coupling capability to run multiple coupled programs in parallel, and an interface to enable Python application coupling. In this paper, we first outline the features of h3-Open-UTIL/MP and its ensemble coupling capabilities, and then report the details of Python application coupling and a coupling case study of the atmospheric model NICAM with a machine learning framework. To couple a Python Application, we added a Python wrapper to the h3-Open-UTIL/MP interface provided as a Fortran module, and made it possible to load Fortran routines as a library in the Python application.. The machine learning framework we used as a coupling example is PyTorch. We targeted the NICAM cloud physics subroutine and trained it to generate output variables from input variables. The results show that there is room for improvement in reproducibility. There can be several ways to improve reproducibility: data selection, improvement of preprocess, and changes in machine learning algorithms. In addition, the machine learning side was the bottleneck of execution time, and the need to improve the computational performance became apparent. One effective way to improve computational performance is to run NICAM on a conventional high performance computer and PyTorch on a GPU machine. We are currently developing a heterogeneous machine coupling program to implement this coupling method, which will be briefly described in the last section.
• Paper (ACM Digital Library)

14:20-14:35 Afternoon Break 1

14:35-16:05 Performance Evaluations

• Session Chairs: Yuan He (primary), Jerry Chou (secondary)

14:35-15:05 "Multi-hetero Acceleration by GPU and FPGA for astrophysics simulation on oneAPI environment"

• Authors: Ryuta Kashino, Ryohei Kobayashi, Norihisa Fujita, and Taisuke Boku (University of Tsukuba)
• Abstract (click to expand):
  GPU computing is one of the most popular accelerating method for various HPC applications. On scientific computations based on multi-physical phenomena, however, a single device solution on GPU is not sufficient where the single time scale or degree of parallelism is not simple to be supported by a simple GPU-only solution. We have been researching a combination of GPU and FPGA coworking toward such complex physical simulation. The most challenging issue is how to program these multiple devices on a single code.
  OneAPI recently provided by Intel is a programming paradigm to support such a solution on a single language platform by DPC++ based on Sycl2020. However, there is no practical application to utilize its full feature to support heterogeneous multi-device programming to demonstrate its potential capability. In this paper, we present the implementation and performance evaluation of our astrophysics code ARGOT to apply oneAPI solution with GPU and FPGA. This work is performed on our multi-heterogeneous accelerated supercomputer Cygnus at University of Tsukuba to realize our concept of CHARM (Cooperative Heterogeneous Acceleration by Reconfigurable Multidevices) for complex multi-physical simulation as a next generation of accelerated supercomputing.
• Paper (ACM Digital Library)

15:05-15:35 "Strong Scaling of OpenACC enabled Nek5000 on several GPU based HPC systems"

• Authors: Jonathan Vincent (KTH Royal Institute of Technology); Jing Gong (Uppsala University); Martin Karp, Adam Peplinski, Niclas Jansson, and Artur Podobas (KTH Royal Institute of Technology); Andreas Jocksch (CSCS - Swiss National Supercomputing Centre); Jie Yao and Fazle Hussain (Texas Tech University); Stefano Markidis (KTH Royal Institute of Technology); Matts Karlsson (Linköping University,); Dirk Pleiter (KTH Royal Institute of Technology); Erwin Laure (Max Planck Computing and Data Facility); and Philipp Schlatter (KTH Royal Institute of Technology)
• Abstract (click to expand):
  We present new results on the strong parallel scaling for the OpenACC-accelerated implementation of the high-order spectral element fluid dynamics solver Nek5000. The test case considered consists of a direct numerical simulation of fully-developed turbulent flow in a straight pipe, at two different Reynolds numbers Re_τ=360 and Re_τ=550, based on friction velocity and pipe radius. The strong scaling is tested on several GPU-enabled HPC systems, including the Swiss Piz Daint system, TACC's Longhorn, Jülich's JUWELS Booster, and Berzelius in Sweden. The performance results show that speed-up between 3-5 can be achieved using the GPU accelerated version compared with the CPU version on these different systems. The run-time for 20 timesteps reduces from 43.5 to 13.2 seconds with increasing the number of GPUs from 64 to 512 for Re_τ=550 case on JUWELS Booster system. This illustrates the GPU accelerated version the potential for high throughput. At the same time, the strong scaling limit is significantly larger for GPUs, at about 2000-5000 elements per rank; compared to about 50-100 for a CPU-rank.
• Paper (ACM Digital Library)

15:35-16:05 "On the Performance Portability of OpenACC, OpenMP, Kokkos and RAJA"

• Author: Ami Marowka (Parallel Reseach Labs)
• Abstract (click to expand):
  Performance Portability frameworks are becoming more central and essential in heterogeneous computing systems. However, the developer toolbox lacks the tools to assess the performance portability degree of these frameworks.
  This article presents a new definition and a metric for evaluating the performance portability of high-level parallel programming models. Using the new metric, the performance portability of OpenACC, OpenMP, Kokkos and RAJA were evaluated based on 324 case studies in various application domains, CPUs and GPUs architectures, and high-performance compilers. The results show that the four performance portability frameworks achieve impressive performance portability of over 80% with no significant differences between different architectures and compilers.
• Paper (ACM Digital Library)

16:05-16:20 Afternoon Break 2

16:20-16:40 Special Session 2

• Session Chairs: Miwako Tsuji (primary), Kento Sato (secondary)

16:20-16:40 "Fugaku, the Successes and the Lessons Learned"

• Speaker: Satoshi Matsuoka (RIKEN R-CCS)

16:40-17:30 Keynote 3

• Session Chairs: Kento Sato (primary), Takaaki Miyajima (secondary)

16:40-17:30 "Swiss National Supercomputing Centre’s Alps as Supercomputing Infrastructure as Code"

• Speaker: Sadaf Alam (Swiss National Supercomputing Centre (CSCS))
• Abstract (click to expand):
  Swiss National Supercomputing Centre (CSCS) has been in the process of developing successor of its current flagship platform called Piz Daint. Like its predecessor, Alps has been co-designed, extending application-driven co-design using accelerator technologies like GPU to Exascale workflows enabling co-design, where we leverage Infrastructure as Code (IaC) technologies. IaC has been a common place in cloud computing, however, it lacked adoption within multi-Petascale ecosystems due to concerns related to performance and interoperability with classic High Performance Computing (HPC) data centers’ ecosystems. This talk will include deployment and configuration details of the first phase of Alps, which is in operations since Q2 2021. Concept of multi-tenant virtual clusters for HPC was introduced by using the Cray-HPE EX Shasta technologies. A similar approach has been adopted for an operational weather forecasting solution for MeteoSwiss at CSCS. Supercomputing IaC delivers close-to-metal performance and scalability for users of supercomputing ecosystems while enabling multi-tenancy for virtual clusters. This allows for programmable configuration of HPC clusters (e.g. Slurm) and cloud clusters (e.g. Kubernetes) such that individual clusters can be managed in isolation and can be provisioned for different quality-of-service (QoS) of customers like MeteoSwiss for 24x7x365 operational forecasts. In addition, programmability aspects of IaC lower or eliminate constraints for the orchestration of workflows incorporating machine learning and experimental data-driven computing, networking and storage services across distributed computing and data management ecosystems.
• Bio (click to expand):
  Sadaf R. Alam is Chief Technology Officer (CTO) at the Swiss National Supercomputing Centre (CSCS) in Lugano, Switzerland. Dr. Alam studied computer science at the University of Edinburgh, UK, where she received her Ph.D. in 2004. Until March 2009, she was a computer scientist at the Oak Ridge National Laboratory, USA. In her role as the CTO, she ensures end-to-end integrity of HPC systems and storage solutions and leads strategic projects at the centre. She has held different roles at CSCS including group lead of future systems, chief architect, and head of operations. She is a member of ACM, ACM-W, SIGHPC and Women in HPC.
• iCalendar

17:30-17:45 Closing (Day 2)

• Kento Sato; Osni Marques

Day 3 (Friday, January 14th)

All times are in JST.

9:00-9:15 Opening (Day3)

• Kento Sato; Osni Marques

9:15-10:45 Architectures and File System

• Session Chairs: M. Mustafa Rafique (primary), Bogdan Nicolae (secondary)

9:15-9:45 "CHFS: Parallel Consistent Hashing File System for Node-local Persistent Memory"

• Authors: Osamu Tatebe, Kazuki Obata, and Kohei Hiraga (University of Tsukuba) and Hiroki Ohtsuji (Fujitsu Limited)
• Abstract (click to expand):
  This paper proposes a design of CHFS, an ad hoc parallel file system using persistent memory of compute nodes. The design is purely based on a highly scalable distributed key value store with consistent hashing. It improves the scalability of the parallel data access performance and metadata performance in terms of the number of compute nodes by not having dedicated metadata servers, any sequential execution and any centralized data management. It is implemented to efficiently utilize multicore and manycore CPUs, high performance networks and remote direct memory access by Mochi-Margo library. In the performance evaluation using a 4-node persistent memory cluster, CHFS shows up to 9.9 times and 6.0 times better performance than the state-of-the-art DAOS distributed object storage and GekkoFS, respectively, in the IOR hard write benchmark. Regarding the scalability, CHFS shows much better scalability and performance than BeeOND and GekkoFS. When using 64 compute nodes, CHFS shows 17.3 and 15.8 times better IO500 bandwidth and metadata score compared to the 1 compute node case, respectively. CHFS shows better scalability and better performance of the metadata and file data access. It would be a promising building block in storage layers in HPC.
• Paper (ACM Digital Library)

9:45-10:15 "A High-Fidelity Flow Solver for Unstructured Meshes on Field-Programmable Gate Arrays: Design, Evaluation, and Future Challenges"

• Authors: Martin Karp and Artur Podobas (KTH Royal Institute of Technology), Tobias Kenter (Paderborn University), Niclas Jansson (KTH Royal Institute of Technology), Christian Plessl (Paderborn University), and Philipp Schlatter and Stefano Markidis (KTH Royal Institute of Technology)
• Abstract (click to expand):
  The impending termination of Moore's law motivates the search for new forms of computing to continue the performance scaling we have grown accustomed to. Among the many emerging Post-Moore computing candidates, perhaps none is as salient as the Field-Programmable Gate Array (FPGA), which offers the means of specializing and customizing the hardware to the computation at hand.
  In this work, we design a custom FPGA-based accelerator for a computational fluid dynamics (CFD) code. Unlike prior work -- which often focuses on accelerating small kernels -- we target the entire Poisson solver on unstructured meshes based on the high-fidelity spectral element method (SEM) used in modern state-of-the-art CFD systems. We model our accelerator using an analytical performance model based on the I/O cost of the algorithm. We empirically evaluate our accelerator on a state-of-the-art Intel Stratix 10 FPGA in terms of performance and power consumption and contrast it against existing solutions on general-purpose processors (CPUs). Finally, we propose a novel data movement-reducing technique where we compute geometric factors on the fly, which yields significant (700+ Gflop/s) single-precision performance and an upwards of 2x reduction in runtime for the local evaluation of the Laplace operator.
  We end the paper by discussing the challenges and opportunities of using reconfigurable architecture in the future, particularly in the light of emerging (not yet available) technologies.
• Paper (ACM Digital Library)

10:15-10:45 "A Method for Reducing Time-to-Solution in Quantum Annealing Through Pausing"

• Authors: Michael Ryan Zielewski and Hiroyuki Takizawa (Tohoku University)
• Abstract (click to expand):
  Recent research has shown that alternative annealing schedules provide the means for improving performance in modern quantum annealing devices. One such type of schedule is forward annealing with a pause, in which there is a period of time when system evolution is paused. While the results from using this type of schedule have been promising, effectively using a pause is not a trivial task. One challenge associated with introducing a pause into the schedule is determining the point in the anneal at which the pause will start. Additionally, tuning the schedule in real-time requires a significant amount of time. A second challenge is that while a pause may increase the number of correct solutions returned from the annealer, the time-to-solution, a standard metric for measuring performance in quantum annealing, will not necessarily be improved. We propose a method for constructing annealing schedules containing a pause that avoids the costly process of determining the optimal pause location in an online manner. We also evaluate our method on the subset sum problem, a problem of practical significance, and show that our method is able to achieve a 70% reduction in time-to-solution from a standard schedule containing no pause.
• Paper (ACM Digital Library)

10:45-11:00 Morning Break

11:00-12:10 Vendor Session (4 talks)

• Session Chairs: Shinji Sumimoto, Takaaki Miyajima

(1) DataDirect Networks, Inc, (DDN)

• Title: "Shared (Posix) File Systems: Past, Present, and Future."
• Speaker: Robert Triendl

(2) Intel

• Title: "Igniting the Next Era of Supercomputing for AII"
• Speaker: Jeff McVeigh

(3) Fujitsu Limited

• Title: "Supercomputer Fugaku and Its Prospects"
• Speaker: Masaru Kase

(4) Oracle Corporation Japan

• Title: "Oracle Cloud Infrastructure (OCI) for HPC"
• Speaker: Tsutomu Miyashita

12:10-12:30 Awarding & Closing

• Session Chair: Miwako Tsuji

12:30-13:00 Lunch Break

13:00-19:10 Workshops

• Attention: Start and end times are different on each workshop.

13:00-16:30 International Workshop on RESource DISaggregation in High Performance Computing - Asia Edition (RESDIS-Asia 2022)

13:30-17:20 IXPUG Workshop HPC Asia 2022

15:00-17:30 International workshop on Arm-based HPC: Practice and Experience (IWAHPCE)

16:00-19:10 Multi-scale, Multi-physics and Coupled Problems on Highly Parallel Systems (MMCP)