Tamito Kajiyama

Monte Carlo code for high spatial resolution ocean color simulations

A Monte Carlo code for ocean color simulations has been developed to model in-water radiometric fields of downward and upward irradiance (E(d) and E(u)), and upwelling radiance (L(u)) in a two-dimensional domain with a high spatial resolution. The efficiency of the code has been optimized by applying state-of-the-art computing solutions, while the accuracy of simulation results has been quantified through benchmark with the widely used Hydrolight code for various values of seawater inherent optical properties and different illumination conditions. Considering a seawater single scattering albedo of 0.9, as well as surface waves of 5 m width and 0.5 m height, the study has shown that the number of photons required to quantify uncertainties induced by wave focusing effects on E(d), E(u), and L(u) data products is of the order of 10(6), 10(9), and 10(10), respectively. On this basis, the effects of sea-surface geometries on radiometric quantities have been investigated for different surface gravity waves. Data products from simulated radiometric profiles have finally been analyzed as a function of the deployment speed and sampling frequency of current free-fall systems in view of providing recommendations to improve measurement protocols.

Performance evaluation of parallel sparse matrix-vector products on SGI Altix3700

The present paper discusses scalable implementations of sparse matrix-vector products, which are crucial for high performance solutions of large-scale linear equations, on a cc-NUMA machine SGI Altix3700. Three storage formats for sparse matrices are evaluated, and scalability is attained by implementations considering the page allocation mechanism of the NUMA machine. Influences of the cache/memory bus architectures on the optimum choice of the storage format are examined, and scalable converters between storage formats shown to facilitate exploitation of storage formats of higher performance.

Regression of in-water radiometric profile data

This study addresses the regression of in-water radiometric profile data with the objective of investigating solutions to minimize uncertainties of derived products like subsurface radiance and irradiance (L(u0) and E(d0)) and diffuse attenuation coefficients. Analyses are conducted using radiometric profiles generated through Monte Carlo simulations and field measurements. A nonlinear NL approach is presented as an alternative to the standard linear method LN. Results indicate that the LN method, relying on log-transformed data, tends to underestimate regression results with respect to NL operating on non-transformed data. The log-transformation is thus identified as the source of biases in data products. Observed differences between LN and NL regression results for L(u0) are of the order of 1-2%, that is well below the target uncertainty for data products from in situ measurements (i.e., 5%). For E(d0), instead, differences can easily exceed 5% as a result of more pronounced light focusing and defocusing effects due to wave perturbations. This work also remarks the importance of applying the multi-cast measurement scheme as a mean to increase the precision of data products.

Performance prediction of ocean color Monte Carlo simulations using multi-layer perceptron neural networks

Abstract A performance modeling method is presented to predict the execution time of a parallel Monte Carlo (MC) radiative transfer simulation code for ocean color applications. The execution time of MC simulations is predicted using a multi-layer perceptron (MLP) neural network regression model trained with past execution time measurements in different execution environments and simulation cases. On the basis of the MLP performance model, a complementary job-environment mapping algorithm enables an efficient utilization of available high-performance computing resources minimizing the total execution time of the simulation jobs distributed in multiple environments.

Scalable software infrastructure project

Recent progress of science and technology has made numerical simulation an important approach for studies in various fields. Although scalable and high performance numerical libraries on large scale computing resources are indispensable tools for handling various multiscale phenomena, few projects for integrating these numerical libraries have been reported. The object of this project is the development of a basic library of solutions and algorithms required for large scale scientific simulations, which have been developed separately in each fields, and its integration into a scalable software infrastructure. The components include a scalable iterative solvers library Lis, having a number of solvers, preconditioners, and matrix storage formats that are flexibly combinable, a fast Fourier transform library FFTSS for various superscalar architectures with SIMD instructions, which outperforms some vendor-provided FFT libraries, and a language- and computing environment-independent matrix computation framework SILC. We show some highlights of our achievements on leading high performance computers.

SILC: a flexible and environment-independent interface for matrix computation libraries

We propose a new framework, named Simple Interface for Library Collections (SILC), that gives users access to matrix computation libraries in a flexible and environment-independent manner. SILC achieves source-level independence between user programs and libraries by (1) separating a function call into data transfer and a request for computation, (2) requesting the computation by means of mathematical expressions in the form of text, and (3) using a separate memory space to carry out library functions independently of the user programs. Using SILC, users can easily access various libraries without any modification of the user programs. This paper describes the design and implementation of SILC based on a client-server architecture, and presents some experimental results on the performance of the implemented system in different computing environments.

High-performance ocean color Monte Carlo simulation in the geo-info project

The Geo-Info project aims to provide Geoscience experts with software toolkits tailored for selected target application domains. Within this framework, high-performance computing (HPC) solutions are here applied to optimize a Monte Carlo (MC) code to support Ocean Color (OC) investigations. The present paper introduces the Geo-Info project, describes the HPC solutions applied for the OC MC case study, and gives early performance results focusing on runtime, speedup, and parallel efficiency.

Distributed SILC: an easy-to-use interface for MPI-based parallel matrix computation libraries

The present paper describes the design and implementation of distributed SILC (Simple Interface for Library Collections) that gives users access to a variety of MPI-based parallel matrix computation libraries in a flexible and environment-independent manner. Distributed SILC allows users to make use of MPI-based parallel matrix computation libraries not only in MPI-based parallel user programs but also in sequential user programs. Since user programs for SILC are free of a source-level dependency on particular libraries and computing environments, users can easily utilize alternative libraries and computing environments without any modification in the user programs. The experimental results of two test problems showed that the implemented SILC system achieved speedups of 2.69 and 7.54 using MPI-based parallel matrix computation libraries with 16 processes.

Statistical Performance Tuning of Parallel Monte Carlo Ocean Color Simulations

Statistical performance tuning of a parallel Monte Carlo (MC) radiative transfer code for ocean color (OC) applications is presented. A low observed-to-peak performance ratio due to highly sparse computations is compensated by online and offline tuning techniques based on a statistical indicator of products accuracy. Run-time adaptive control employs the accuracy indicator to set up two complementary tuning criteria: one general to MC computations and the other specific to OC applications. The same accuracy indicator is also used for pre-execution tuning of a threshold parameter. Numerical simulations of real case scenarios showed that the proposed methods consistently led to faster runs, while satisfying application accuracy requirements. Specifically, speed-ups range from 2.17 to 7.44 times when compared with the un-optimized version of the MC code. The applied techniques are orthogonal to parallelization, so that the reported performance gains are further amplified by parallel speed-ups.

Cloth simulation in the SILC matrix computation framework: a case study

This paper presents a case study of numerical simulations in an easy-to-use matrix computation framework named Simple Interface for Library Collections (SILC), which allows users to use matrix computation libraries in an environment- and language-independent manner. As a practical example of numerical simulations in SILC, we selected cloth simulation based on a mass-spring model and the implicit backward Euler method. We constructed two SILC-based versions of an existing cloth simulation code according to two proposed application styles of SILC. Experimental results showed that both versions achieved some performance gains, thereby demonstrating the feasibility of numerical simulations in SILC and the usability of the proposed application styles.