Nesting Ocean Model for parallel vector processors Miyazawa

  • Slides: 16
Download presentation
Nesting Ocean Model for parallel vector processors Miyazawa, Y. (IGCR/FRSGC) et al. Prediction and

Nesting Ocean Model for parallel vector processors Miyazawa, Y. (IGCR/FRSGC) et al. Prediction and estimation of effects on coastal ocean of climate variation ----- abnormal sea level, Kyucho Prediction of maritime disaster ----- oil spill High-resolution coastal ocean model nested in a basin-scale ocean model

Contents 1. Nesting Model for Kuroshio simulation 2. The Earth Simulator 3. Modification of

Contents 1. Nesting Model for Kuroshio simulation 2. The Earth Simulator 3. Modification of the model code for the Earth Simulator

Nesting POM driven by the NCAR daily data Highresolution model 1/12 deg. (10 km)

Nesting POM driven by the NCAR daily data Highresolution model 1/12 deg. (10 km) Lowresolution model 1/4 deg. (30 km)

Nesting grid One-way nesting Bilinear interpolation 3 x 3 grids

Nesting grid One-way nesting Bilinear interpolation 3 x 3 grids

Eddy kinetic energy

Eddy kinetic energy

Prediction of Kuroshio path (initialized by DA of T/P in 1993. 01) Observation (1993.

Prediction of Kuroshio path (initialized by DA of T/P in 1993. 01) Observation (1993. 03 -1993. 04) Simulation

Nesting of another coastal model (by a TIT group) Part of the Kuroshio intrudes

Nesting of another coastal model (by a TIT group) Part of the Kuroshio intrudes the Tokyo Bay 1/48 deg. (2. 5 km) grid

Earth Simulator Developed by JAERI/JAMSTEC/NASDA 8 vector processors(8 x 8 Gflops) share main memory(16

Earth Simulator Developed by JAERI/JAMSTEC/NASDA 8 vector processors(8 x 8 Gflops) share main memory(16 GB) 1 node 640 nodes are connected by a single-stage crossbar network

Earth Simulator

Earth Simulator

Earth Simulator

Earth Simulator

Speed up of the POM code • Vectorization: 99. 7 % • Parallelization by

Speed up of the POM code • Vectorization: 99. 7 % • Parallelization by MPI (SPMD) • domain decomposition method ----- latitudinal direction • initialization for MPI and subroutines for communication are inserted in the POM

Efficiency of the code Low-resolution model: 543 x 282 x 21 On NEC/SX-5

Efficiency of the code Low-resolution model: 543 x 282 x 21 On NEC/SX-5

Efficiency of the code High-resolution model: 758 x 530 x 35 On NEC/SX-5

Efficiency of the code High-resolution model: 758 x 530 x 35 On NEC/SX-5

Nesting by MPI Low-resolution model High-resolution model mkbnd set_openbnd send_nestbuf recv_nestbuf MPI Network

Nesting by MPI Low-resolution model High-resolution model mkbnd set_openbnd send_nestbuf recv_nestbuf MPI Network

Problems • More tests for the nesting ocean model • Asynchronous coding of MPI

Problems • More tests for the nesting ocean model • Asynchronous coding of MPI • Development of higher-resolution ocean models for Earth Simulator

Another POM codes for Earth Simulator ES-OGCM (by T. Kagimoto) • • Global Model

Another POM codes for Earth Simulator ES-OGCM (by T. Kagimoto) • • Global Model Atmosphere/Ice/Ocean Couple Model Fortran 90/95 Parallelization by MPI: asyncronous communication