RIKEN Center for Computational Science

Menu
Menu
Events/Documents イベント・広報

R-CCS Cafe

R-CCS Cafe is a place where R-CCS researchers can informally discuss their research beyond the boundary of their discipline to facilitate integration of different disciplines. R-CCS Cafe is held twice a month. All who are interested are welcome to attend.

  • Purpose: To provide a forum for researchers to exchange ideas and information, with the goal to facilitate interdisciplinary collaboration and develop new research fields.
  • Place: Lecture Hall (6th floor) or Seminar Room (1st floor) at R-CCS
  • Language: Presentations will be in Japanese or English. Slides will be in English.

Please make your presentation understandable to researchers in other fields. Questions and active discussion are encouraged.

The 150th R-CCS Cafe-part Ⅲ
Date and Time: Fri. Oct. 19, 2018, 15:30 - 16:30
Place: Lecture Hall (6th floor) at R-CCS

Title: 職域救命説明会
Speaker: Ryoko Kawamura (Health Care Center)

Presentation Language: Japanese
Presentation Material: Japanese

Abstract: Detail

R-CCSでは、年間多くのセミナーが各チーム主催で開催されます。また『京』コンピューターの見学のために年間約1万人の外来者が来場しています。もし、主催するセミナーで心肺停止の方が発生した場合はどのように対応しますか?救急通報を行ってから、8分間何も行わなかった場合は救命の可能性は10%を切ってしまいます。救命の説明と昨年度神戸市市民救命講習を受講された方によるデモンストレーションでR-CCSにおける救命の手順を知るきっかけとしてください。また、今回救急時に使用するために購入したフルフラットになる車いすの取扱い説明も行います。

The 149th R-CCS Cafe-part I
Date and Time: Wed. Oct. 10, 2018, 13:45 - 14:25
Place: Lecture Hall (6th floor) at R-CCS

Title: Filter consists of a few resolvents to solve symmetric definite generalized eigenproblems
Speaker: Hiroshi Murakami (Tokyo Metropolitan University)

Presentation Language: English
Presentation Material: English

Abstract: Detail

By using a filter, those eigenpairs of a symmetric definite generalized eigenproblem "A v = lambda Bv" are solved whose eigenvalues are in a specified real interval "[a,b]". In present study, the filter we use is a polynomial of the real part of a linear combination of a few resolvents. Applications of a few resolvents are given by solving corresponding systems of linear equations. We are to solve these systems of linear equations by some matrix decomposition methods. Since we use a few resolvents, the number of matrix decompositions required is also a few (2 to 4).

The 149th R-CCS Cafe-part II
Date and Time: Wed. Oct. 10, 2018, 14:30 - 15:10
Place: Lecture Hall (6th floor) at R-CCS

Title: 1. The BEAST eigensolver and 2. Recent developments and results for the BEAST eigensolver
Speaker: Martin Galgon (University of Wuppertal), Sarah Huber (University of Wuppertal)

Presentation Language: English
Presentation Material: English

Abstract: Detail

1. We will introduce the BEAST eigensolver, which incorporates both polynomial and contour integral type filter diagonalization schemes for the solution of large sparse eigenproblems. The solver incorporates multiple levels of parallelism as well as a variety of algorithmic advances for fast, accurate, and robust solution. The underlying filter diagonalization schemes and the resulting solver will be introduced here.
2. In this talk we will discuss some of the latest advances of the BEAST eigensolver, including the availability of mixed precision calculations for reduced calculation time while maintaining accuracy. We also show results from recent tests showing the scalability of BEAST to very large problem sizes and numbers of cores.

The 149th R-CCS Cafe-part III
Date and Time: Wed. Oct. 10, 2018, 15:50 - 16:20
Place: Lecture Hall (6th floor) at R-CCS

Title: High-performance implementation of the Jacobi based eigen- and singular value decomposition methods
Speaker: Shuhei Kudo (Large-scale Parallel Numerical Computing Technology Research Team)

Presentation Language: English
Presentation Material: English

Abstract: Detail

In this talk, we explain our progress in developing high-performance implementations of the Jacobi method. The Jacobi method is an old algorithm to compute eigen- and singular value decompositions. Although its large computational cost, it has advantages against the conventional method on recent supercomputers when it is used with the blocking and the parallelizing techniques which improve the arithmetic intensity and reduce the communication cost. Our optimized implementations of the Jacobi method show high strong scalability on recent supercomputers with more than ten thousand nodes and outperform the conventional method.

The 149th R-CCS Cafe-part IV
Date and Time: Wed. Oct. 10, 2018, 16:25 - 16:55
Place: Lecture Hall (6th floor) at R-CCS

Title: Numerical evaluation of preconditioned CG method with higher precision
Speaker: Satoshi Kawai (HPC Usability Research Team)

Presentation Language: English
Presentation Material: English

Abstract: Detail

Solving large-scale systems of linear equations which has sparse coefficient matrices derived from generalized eigenvalue problems are challenging for iterative solvers. Although the coefficient matrices have ill-conditioned, eigenvalue solvers require higher accuracy of approximate solutions. In this study, we evaluate an effect of a higher precision of a Krylov subspace method with a parallel ILU preconditioner. The ILU preconditioner with regularizations and multi-coloring parallelism showed higher robustness and good performance. To achieve higher accuracy and convergence, we applied quad precision to an IC-COCG method. In this talk, we will have a discussion about the accuracy, convergence ratio and computational time among single, double and quad precisions of the IC-COCG method with well and ill-conditioned problems.

The 148th R-CCS Cafe-part Ⅰ
Date and Time: Fri. Oct. 5, 2018, 13:00 – 14:00
Place: Lecture Hall (6th floor) at R-CCS

Title: Toward a Science of HPC software (and hardware)
Speaker: Junichiro Makino (Team Leader, Particle Simulator Research Team)

Presentation Language: English
Presentation Material: English

Abstract: Detail

In the last three decades, it has been becoming more and more difficult to achieve high, or even modest efficiency on advanced HPC systems. As everybody here in R-CCS knows, the development of highly efficient and highly scalable software for modern HPC systems like K or Post-K requires multi-year effort of a team of exceptionally good researchers. Of course, many other HPC systems are even harder to use. Our particle simulator research team has been working on a new approach to solve this problem through the use of framework. Our FDPS framework allows the users to write their programs in the way very close to the way they would write them for a single-processor machine, and run them on large-scale parallel systems with very high efficiency. The guiding principle for the design of FDPS is to minimize inter-node communication and main memory access per timestep, without increasing the floating-point operations too much. I will discuss current state of FDPS and also our framework for stencil calculation, Formura. If time allows, I'll discuss how hardware design should be changed to achieve better efficiency, not just that relative to peak but the power efficiency which is far more important.

The 148th R-CCS Cafe-part Ⅱ
Date and Time: Fri. Oct. 5, 2018, 14:15 – 15:15
Place: Lecture Hall (6th floor) at R-CCS

Title: An Overview of Post-K Hardware and System Software
Speaker: Yutaka Ishikawa (Project Leader, Flagship 2020 Project)

Presentation Language: English
Presentation Material: English

Abstract: Detail

RIKEN R-CCS and Fujitsu are developing Post-K, successor of K computer, that will be available for early access program in Q2 of 2020 and will begin general operation in 2021. Fujitsu recently announced the specification of CPU, named A64FX, and its interconect, TofuD, for Post-K. A64FX is the first implementation of Arm 64bit HPC architecture (Scalable Vector Extension) and TofuD is an enhanced version of Tofu used in K computer. The system software developed in the post K supercomputer includes a novel operating system for general-purpose manycore architectures, low-level communication and MPI libraries, and file I/O middleware. In this presentation, after inroducing an overview of Post-K hardware and system software, future research directions will be addressed.

The 147th R-CCS Cafe
Date and Time: Fri. Sep. 28, 2018, 15:30 – 16:30
Place: Lecture Hall (6th floor) at R-CCS

Title: Integral equation theories of diffusion for molecular liquids
Speaker: Kento Kasahara (Laboratory for Biomolecular Function Simulation (BDR) / Computational Biophysics Research Team (R-CCS))

Presentation Language: English
Presentation Material: English

Abstract: Detail

Understanding the dynamical behavior of liquids is a main subject of solution chemistry. Liquid phase is a complicated many-body system and its dynamics can not be understand only with tracing the motion of each molecule. Diffusion equation can be regarded as a simple model for characterizing such many-body dynamics in the long time regime. However, the equation does not consider the intermolecular interactions and molecular geometries, meaning that the equation can not be used to understand the phenomena at the microscopic level. In order to overcome the problems, dynamics theories for molecular liquids have been developed over the past few decades. The theories are based on reference interaction site model (RISM), an integral equation theory for molecular liquids, and Zwanzig-Mori projection operator method. Recently, we developed a molecular diffusion theory which is applicable to complicated systems such as electrolyte solutions for lithium-ion batteries, and constructed a fundament to investigate diffusion-controlled reactions in polyatomic molecule systems. In the theories, molecular diffusion is described as the time-evolution of the spatial distribution function for each solvent atom (interaction site) around a solute molecule. The function is obtained by solving the formulated equations analytically or numerically, and hence the theories are free from statistical error which sometimes appear as severe problems in simulation-based methods. In this talk, I’ll present our approaches in molecular diffusion based on statistical mechanics, together with some applications. At the end of the talk, I’ll briefly introduce the current research project in Sugita group (BDR), ligand binding process in a crowded environment.

The 146th R-CCS Cafe-part Ⅰ
Date and Time: Fri. Sep. 7, 2018, 13:00 – 14:00
Place: Lecture Hall (6th floor) at R-CCS

Title: Review of our climate studies and future direction
Speaker: Hirofumi Tomita (Team Leader, Computational Climate Science Research Team)

Presentation Language: English
Presentation Material: English

Abstract: Detail

Climate study has two aspects: basic understanding of physical mechanism and future climate projection. In the former study, how and why the equilibrium state with fluctuation and drastic regime shift do occur are investigated. On the other hand, the latter study focusses on the near-future climate assessment and directly contributes to the society by providing the reliable information. For proceeding the climate study, it is necessary to understand several important but basic feedbacks accompanying with greenhouse gases, e.g., carbon cycle feedback, ice-albedo feedback, cloud feedback, water vapor feedback, and so on. As an introduction, I will give explanation for such processes as simply as possible.
Although the climate study relies on the global/regional model, there exists several issues in climate modeling, numerical convergence at high resolution, cloud expression, turbulence theory, and so on. I will review our representative climate studies in the last several years: numerical convergence of deep convection and prediction skill of Madden Julian Oscillation, poleward transport of black carbon, and indirect effect of aerosol on cloud in the global model study, and numerical exploration of dust devil aspect in Mars and the development of a new assessment method of regional climate in the regional model study.
I also talk about the future direction of our research. As the modeling issue, the turbulence-cloud coupling scheme is a crucial issue. This would require reconstruction of turbulence theory including water vapor saturation. Scientifically, cloud organization and hierarchical structure are a key for understanding the climate equilibrium when such phenomena are regarded as the energy system.

The 146th R-CCS Cafe-part Ⅱ
Date and Time: Fri. Sep. 7, 2018, 14:00 – 15:00
Place: Lecture Hall (6th floor) at R-CCS

Title: Programming for Task parallelism – A new trend for manycore programming
Speaker: Mitsuhisa Sato (Team Leader, Programming Environment Research Team)

Presentation Language: English
Presentation Material: English

Abstract: Detail

Task programming has been newly introduced in OpenMP and is attracting much attention to make use of manycore efficiently. Task in OpenMP is a code block which can be executed in parallel. In OpenMP 4.0, task dependency is defined to describe data dependency between tasks. It enables to exploit parallelism in a different way from conventional OpenMP construct. In this talk, a short introductory tutorial on programming for task parallelism will be presented, and is followed by research to extend the task programming to distributed memory environment including XcalalableMP 2.0 and integration with PGAS.