Software requirements
To follow this tutorial, first ensure that:
You will be working on a Linux platform. Use of the Windows or Mac OS X remote client to connect to a Linux host is not part of this tutorial (see Connecting to a remote system for more information).
You have Linaro Forge installed. See Linaro Forge Downloads and Installing Linaro Forge.
Your Linaro Forge installation is licensed. Copy your license file into the
licencesdirectory of your Linaro Forge installation.Your Linaro Forge license includes the MPI Comm Matrix feature. Contact Linaro Sales for details about how to upgrade.
A Message Passing Interface (MPI) environment is configured. If you are unsure if you have an MPI environment configured, ask your System Administrator.
You have access to a multi-core or a multi-node machine. This worked example has been tested and tuned on a single-node system, using 16 cores.
To follow along with this tutorial, you will need the proxy app source files. We have provided three versions of the application, representing the three stages of our optimization journey.
fire_baseline.cpp: The naive implementation (1D Block +MPI_Gatherv).
fire_io.cpp: The I/O optimized version (1D Block + Parallel MPI-IO).
fire_balanced.cpp: The compute optimized version (Static Load Balancing + Parallel MPI-IO).
You may want to copy those files over into a newly-created directory, as running them will result in generation of artifacts, such as a binary file containing snapshots of fire propagation. You may also want to edit files to perform described optimizations manually (even though we provide optimized versions), or try your own further optimizations.
A Python visualization script is also provided in the repository to read the resulting wildfire.dat
file and render the fire’s progression as an animation (this is optional).