Pāriet uz saturu

Translation status

This page has not yet been translated into Latvian, or its translation is out of date. The current English version is shown as a fallback.

FrontISTR v4.3 Release Notes

Introduction to the Minor Version Upgrade (VER.4.3)

PExProCS, LLC Kazuya Goto

Table of Contents

  • Overview of the minor version upgrade
  • Added features
    • Analysis function related
    • Partitioner related
    • Mesh / refine related
    • Linear solver related
  • Bug fixes, and others
  • Scheduled release

Overview of the Minor Version Upgrade

  • Incorporation of code that existed in the development group but was not included in the public version
  • Incorporation of function extensions and customizations developed in the Okuda Laboratory and its surroundings (including joint research)
  • Implementation of (a part of) the features requested by users
  • Fixing of defects
  • Support for mixed analysis of shell/beam elements and solid elements
    • (Details on the next page)
  • Support for orthotropic materials in shell elements
  • Support for layered shells
  • Load FOLLOW function for large-deformation analysis
    • Added the parameter FOLLOW to !DLOAD (enabled by default)
  • Support for beam elements in dynamic analysis
    • There are restrictions on the applicable load types (see the manual)
  • Monitoring of multiple nodes in dynamic analysis
    • Specification by node group is possible
  • Monitoring of nodal stress / nodal strain in dynamic analysis
  • Application of window functions to the input fluid force in coupled analysis

Mixed Analysis of Shell/Beam Elements and Solid Elements

  • Added beam elements (element type 641) that assign 3 degrees of freedom per node to 2 nodes, and shell elements (element types 761, 781)
  • Since this is an experimental introduction, there is a possibility of specification changes in the future

Element type: 641
Nodes 1 and 2 have translational degrees of freedom, and nodes 3 and 4 have rotational degrees of freedom

Element type: 761
Nodes 1, 2, and 3 have translational degrees of freedom
Nodes 4, 5, and 6 have rotational degrees of freedom

Element type: 781
Nodes 1, 2, 3, and 4 have translational degrees of freedom
Nodes 5, 6, 7, and 8 have rotational degrees of freedom

v4.3 V4.3_140221.pdf p.5 original figure

  • Substantial speedup
    • A speedup of about 150x in the case of partitioning a model of about 40 million nodes into about 20,000 domains
  • Support for Metis Ver.5
    • Continued support for Ver.4 as well
    • When using MUMPS, caution is required because MUMPS supports only Metis Ver.4
  • Support for refinement of mixed element type models
    • However, mixing of element types with different nodal degrees of freedom is not possible
    • Support for refinement of plane models, and of shell/beam models, as well
  • Support for interpolation based on the refinement information for input temperature data
    • For example, performing a heat conduction analysis without refinement, or with 1 refinement, and using its result as input to perform a thermal stress analysis with 2 refinements, etc.
  • Multi-color processing and hybrid parallel support in preconditioning
    • Through the cooperation of the Research Institute for Advanced Science and Technology (RIST)
    • (Details on the next page)
  • Ordering for vector computers
    • Through the cooperation of the Japan Agency for Marine-Earth Science and Technology (JAMSTEC)
    • Added the parameter USEJAD to !SOLVER
  • For 3-degree-of-freedom problems, all combinations of preconditioners and iterative solvers are available
    • Iterative solvers: CG, BiCGSTAB, GPBiCG, GMRES
    • Preconditioners: SSOR, diagonal scaling, BILU(0), BILU(1), BILU(2)
  • Matrix data dump function
    • Mainly for debugging
    • Added the parameters DUMPTYPE and DUMPEXIT to !SOLVER

Multi-color Processing and Hybrid Parallelism in Preconditioning

  • Parallelized the matrix-vector product in threads using OpenMP
  • Introduced multi-color ordering into the SSOR preconditioner and parallelized it in threads using OpenMP
  • Tuned for the K computer and FX10

Figure: Thread parallel performance of the linear solver portion on the K computer (Data is from tutorial/02_elastic_hinge_parallel (4 domains))

v4.3 V4.3_140221.pdf p.9 original figure

Bug Fixes, and Others

  • Fix of the Drilling DOF of shell elements
  • Fix of the convergence criterion in nonlinear static analysis
  • Fix of the time at restart in linear dynamic analysis
  • Fix of the node number display in messages during coupled analysis
  • Fix of the ILU preconditioner
  • Fix of the communication routines of HEC-MW
  • Avoiding defects during optimization by some compilers, etc.

Scheduled Release of Ver. 4.3

  • Release timing: February 2014
  • Version name: Ver. 4.3
  • Scheduled release of the binary package for Linux
    • rpm (for Redhat-family distributions)
    • deb (for Debian-family distributions)
  • Public site
    • http://www.multi.k.u-tokyo.ac.jp/FrontISTR/