Keiichiro Fukazawa

Analyzing and mitigating the impact of manufacturing variability in power-constrained supercomputing

A key challenge in next-generation supercomputing is to effectively schedule limited power resources. Modern processors suffer from increasingly large power variations due to the chip manufacturing process. These variations lead to power inhomogeneity in current systems and manifest into performance inhomogeneity in power constrained environments, drastically limiting supercomputing performance. We present a first-of-its-kind study on manufacturing variability on four production HPC systems spanning four microarchitectures, analyze its impact on HPC applications, and propose a novel variation-aware power budgeting scheme to maximize effective application performance. Our low-cost and scalable budgeting algorithm strives to achieve performance homogeneity under a power constraint by deriving application-specific, module-level power allocations. Experimental results using a 1,920 socket system show up to 5.4X speedup, with an average speedup of 1.8X across all benchmarks when compared to a variation-unaware power allocation scheme.

Full electromagnetic Vlasov code simulation of the Kelvin-Helmholtz instability

Recent advancement in numerical techniques for Vlasov simulations and their application to cross-scale coupling in the plasma universe are discussed. Magnetohydrodynamic (MHD) simulations are now widely used for numerical modeling of global and macroscopic phenomena. In the framework of the MHD approximation, however, diffusion coefficients such as resistivity and adiabatic index are given from empirical models. Thus there are recent attempts to understand first-principle kinetic processes in macroscopic phenomena, such as magnetic reconnection and the Kelvin–Helmholtz (KH) instability via full kinetic particle-in-cell and Vlasov codes. In the present study, a benchmark test for a new four-dimensional full electromagnetic Vlasov code is performed. First, the computational speed of the Vlasov code is measured and a linear performance scaling is obtained on a massively parallel supercomputer with more than 12 000 cores. Second, a first-principle Vlasov simulation of the KH instability is performed in order ...

Vlasov simulation of the interaction between the solar wind and a dielectric body

The global structure of wake field behind an unmagnetized object in the solar wind is studied by means of a 2.5-dimensional full-electromagnetic Vlasov simulation. The interaction of a plasma flow with an unmagnetized object is quite different from that with a magnetized object such as the Earth. Due to the absence of the global magnetic field, the unmagnetized object absorbs plasma particles that reach the surface, generating a plasma cavity called “wake” on the antisolar side of the object. For numerical simulations of electromagnetic structures around the wake, it is important to include the charging effect in global-scale simulations. The present study is one of the first attempts to study the formation of wake fields via a full-kinetic Vlasov simulation. It has been confirmed that the spatial structures of wake fields depend on the direction of interplanetary magnetic fields as well as the distance from the body.

A Scalable Full-Electromagnetic Vlasov Solver for Cross-Scale Coupling in Space Plasma

Numerical schemes for solving the Vlasov-Maxwell system equations are studied. For studies of cross-scale coupling between fluid-scale dynamics and particle-scale kinetics in collisionless plasma, both highly scalable kinetic code and huge supercomputer are essential. In the present study, a new parallel Vlasov-Maxwell solver is developed by adopting a stable but less dissipative scheme for time integration of conservation laws. The new code has successfully achieved a high scalability on massively parallel supercomputers with multicore scalar processors. The new code has been applied to 2P3V (two dimensions for position and three dimensions for velocity) problems of cross-scale plasma processes such as magnetic reconnection, Kelvin-Helmholtz instability, and interaction between the solar wind and an asteroid.

Power Consumption Evaluation of an MHD Simulation with CPU Power Capping

Recently to achieve the Exa-flops next generation computer system, the power consumption becomes the important issue. On the other hand, the power consumption character of application program is not so considered now. In this study we examine the power character of our Magneto hydrodynamic (MHD) simulation code for the global magnetosphere to evaluate the power consumption behavior of the simulation code under the CPU power capping on the parallel computer system. As a result, it is confirmed that there are different power consumption parts in the MHD simulation code, which the execution performance decreases or does not change under the CPU power capping. This indicates the capability of performance optimization with the power capping.

Performance Measurement of Magnetohydrodynamic Code for Space Plasma on the Various Scalar-Type Supercomputer Systems

The computational performance of magnetohydrodynamic (MHD) code is evaluated on several scalar-type supercomputer systems. We have made performance tuning of a 3-D MHD code for space plasma simulations on the SR16000/L2, FX1, and HX600 supercomputer systems. For parallelization of the MHD code, we use four different methods, i.e., regular 1-D, 2-D, and 3-D domain decomposition methods and a cache-hit type of 3-D domain decomposition method. We found that the regular 3-D decomposition of the MHD model is suitable for HX600 system, and the cache-hit type of 3-D decomposition is suitable for SR16000/L2 and FX1 systems. As results of these runs, we achieved a performance efficiency of almost 20% for MHD code on all systems.

The HLLD Approximate Riemann Solver for Magnetospheric Simulation

A magnetohydrodynamic (MHD) algorithm for global simulations of planetary magnetospheres is developed based on an approximate nonlinear Riemann solver, the so-called Harten-Lax-van Leer-Discontinuities (HLLD) approximate Riemann solver. An approximate nonlinear solution of the MHD Riemann problem, in which the contributions of the background potential magnetic field are subtracted and multispecies plasmas as well as general equation of state are included, can be algebraically obtained under the assumptions that the normal velocity and the background potential magnetic field in the Riemann fan are constant. The theoretical aspects of the HLLD approximate Riemann solver are focused on, in particular.

The Locations and Shapes of Jupiter's Bow Shock and Magnetopause

The shape and location of the Jovian bow shock and magnetopause have been studied by using magnetic field observations and global magnetohydrodynamic (MHD) simulations. MHD simulations in which the interplanetary magnetic field (IMF) was set to zero were used to define the boundary shapes and positions and how they depend on solar wind dynamic pressure. Polynomial fits to the simulated boundaries along with spacecraft observations were used to determine the probability of a given position being outside of the bow shock or inside of the magnetopause. The magnetopause and possibly the bow shock have two preferred locations, one representing a compressed magnetosphere and the other an expanded magnetosphere. Variations in the solar wind parameters near Jupiter also show a bimodal distribution but the changes in the solar wind dynamic pressure are not sufficient to account for the observed bimodal distribution of observed magnetopause positions. Internal pressure changes at Jupiter are required. The interplanetary magnetic field also influences the location and shape of the boundaries. In particular, when the IMF is in the B y direction or northward magnetopause reconnection acts to reduce polar flattening. Higher internal pressure at dusk leads to a dawn-dusk asymmetry in the magnetopause position with the boundary being farther from Jupiter on the dusk side. For all the simulations the ratio of the bow shock stand-off distance to that of the magnetopause was less than that at the Earth.

A high-resolution global Vlasov simulation of a small dielectric body with a weak intrinsic magnetic field on the K computer

The interaction between the solar wind and solar system bodies, such as planets, satellites, and asteroids, is one of the fundamental global-scale phenomena in space plasma physics. In the present study, the electromagnetic environment around a small dielectric body with a weak intrinsic magnetic field is studied by means of a first-principle kinetic plasma simulation, which is a challenging task in space plasma physics as well as high-performance computing. Due to several computational limitations, five-dimensional full electromagnetic Vlasov simulations with two configuration space and three velocity space coordinates are performed with two different spatial resolutions. The Debye-scale charge separation is not solved correctly in the simulation run with a low spatial resolution, while all the physical processes in collisionless plasma are included in the simulation run with a high spatial resolution. The direction comparison of electromagnetic fields between the two runs shows that there is small difference in the structure of magnetic field lines. On the other hand, small-scale fine structures of electrostatic fields are enhanced by the electric charge separation and the charge accumulation on the surface of the body in the high-resolution run, while these structures are absent in the low-resolution runs. These results are consistent with the conventional understanding of plasma physics that the structure and dynamics of global magnetic fields, which are generally described by the magneto-hydro-dynamics (MHD) equations, are not affected by electron-scale microphysics.

Performance Tuning of Vlasov Code for Space Plasma on the K Computer

Space plasma is a collisionless, multi-scale, and highly nonlinear medium. Thus computer simulations are essential for full understanding of space plasma. In the present study, we develop a high-performance parallel Vlasov (collisionless Boltzmann) simulation code which is the first-principle method for collisionless space plasma. The performance tuning of the code has been made on various supercomputer systems such as the K computer, FX10 and CX400 supercomputer systems. The performance efficiency of more than 15% is achieved on these systems.