Adaptive computational fluid dynamics
Inciter: Partial differential equations solver on 3D unstructured grids

Inciter is a fully asynchronous distributed-memory-parallel finite element field solver for partial differential equations using 3D unstructured grids. It is used to research asynchronous mesh-based algorithms using overdecomposition and to experiment with coupling asynchronous and bulk-synchronous parallel code using widely used libraries.

Inciter uses the Charm++ runtime system to run concurrently, either on a single machine or a networked set of computers. The software design is fully asynchronous and it exhibits perfect, i.e., 100% efficiency, strong scaling.

Help on command-line parameters

As usual, use the -h command-line parameter to get on-screen help from an executable. Example output, excluding Charm++ command-line parameters:

$ inciter -h
inciter Command-line Parameters:
-c, --control string Specify the control file name [REQUIRED]
-h, --help Display one-liner help on all command-line arguments
-f, --helpctr Display one-liner help on all control file keywords
-H, --helpkw string Display verbose help on a single keyword
-i, --input string Specify the input file
-o, --output string Specify the output file
-v, --verbose Select verbose screen output
-u, --virtualization real Set degree of virtualization

Equations solved

Currently, Inciter can solve a system of linear advection-diffusion equations. It has parallel I/O, graph-based domain decomposition using Zoltan2, asynchronous linear system assembly, and solves the distributed linear system using hypre. The figure on the right quantifies the excellent scalability of this prototype, taking 10 time steps with a 76M-cell tetrahedron mesh including setup and I/O, up to 4K CPU cores, the most tested. The insert depicts the CPU utilization from Charm++'s performance analysis tool, Projections, showing excellent resource usage during setup (left side) as well as during the 10 two-stage time-steps.

Page last updated: Thu 06 Apr 2017 10:44:07 AM MDT Copyright 2012-2015, J. Bakosi, 2016-2018, Los Alamos National Security, LLC.