This section describes technologies implemented only in FlowVision 3.xx.xx (FlowVision-HPC).

Modular architecture

Distributed client-server architecture

• Components:
  • Terminal: Linux/Windows
  • Pre-Postprocessor: Windows
  • Solver and Solver-Agent: Linux/Windows
  • License Manager: Linux/Windows
  • Viewer: Linux/Windows
  • Multi-Physics Manager, MPM-Agent, MPM Connector: Linux/Windows
• Client applications have user interface
• Server applications don't have user interface
• License Manager can be installed on any machine assessable to all the FlowVision components through the local network
• Solver-Agent and Solver are always automatically installed on one machine
• MPM-Agent has to be installed on the machine where Abaqus is installed
• Typical configurations
  • One Windows machine
  • Calculations on separate machine connected to client machine
  • Calculations on separate machine not connected to client machine
  • Cluster

High performance computing

• Data exchange between processors: Message Passing Interface (MPI)
• Automatic decomposition of computation domain onto processors
• High scalability of computations on computers with shared, distributed, and combined memory
• All the computation processes are parallelized, viz.,
  • decomposition of the computation domain
  • building the computational grid
  • iterative solving the systems of algebraic equations
  • processing and visualizing results

Decomposition of computation domain onto processors

Decomposition of grid around golf ball onto several processors
by buil-in algorithm

• Parallel volume decomposition of the computation domain: ParMETIS
• Parallel decomposition of the interprocessor boundaries: METIS
• Decomposition processes are automated
• The built-in algorithm makes the computational loads on the processors uniform
• You can toggle off the algorithm and specify Cartesian orts along which you want to decompose the computation domain

Parallel visualization

• Computations required for building color contours, isolines, plots, isosurfaces, etc. are performed on several processors
• Minimum data volume is collected on the client machine
• Possibility for several users to simultaneously access solution from different machines