Use Parallel Processing with Wind-US to Accelerate Your Solutions

Parallel processing has become a necessary tool for almost all production CFD work. For complex configurations, you have to run in parallel in order to get timely results. Not everyone can afford a supercomputing system, however, so it's nice that Wind-US has a lot of flexibility in how it can be run in parallel.

Masters and Slaves

When the Wind-US CFD solver is run in parallel, the various blocks in a multi-block grid are parceled out among the available processors (illustrated below). One processor acts as the “master,” and the others are the “slaves.”

Schematic of parallel processing with the Wind-US CFD solver.

Parallel Processing for the Masses

Wind-US has been programmed with two different parallel processing technologies: MPI and the older PVM. On multi-processor systems, Wind-US can be run using either of them to take advantage of the faster interprocessor communication.

Most large systems these days are some variation of the Linux “Beowulf” concept, with hundreds or thousands of separate boxes clustered together. On such a system, you will probably want to use MPI for parallel operation. This is especially true if there is a high-speed interconnect, because MPI is usually the only way to access it.

PVM can even be used to run Wind-US on a combination of heterogeneous systems. As long as you have an executable that is compatible with each system, you can link together Intel systems, AMD-based boxes, Sun SPARC-based systems...almost anything. Wind-US has even been made to run on Microsoft Windows systems (although the NPARC Alliance does not itself currently provide support for Windows systems).

Even More Options

When running in parallel, the “slave” processors don't even need to have access to the file system with the grid and run-control files, as long as the “master” processor does. Of course, if some sort of “global” file system is used such that all the processors can “see” the same files, then Wind-US can take advantage of that for faster operation.

Recent improvements to the parallel algorithms in Wind-US have resulted in quite good performance when run in parallel, even if each processor only works on a few thousand points.

Is This What You Need?

Because of this flexibility, almost any organization can use parallel processing for their CFD...no matter how shoestring the budget. I've set up Wind-US for parallel operations on everything from a basic Linux PC in a home office to massive clusters with thousands of processors. If you would like, you can contact me about parallel processing with Wind-US, and we can go over the possibilities for applying it in your organization.

The next section of this review goes over the pre-processors which are shipped with Wind-US to help you prepare your grids for use with the solver.

Or, head back to the introduction page for this review of Wind-US.

If you would prefer, head over to the free CFD solvers page or the Innovative CFD home page and look at some of the other material.

New! Comments

Have your say about what you just read! Leave a comment in the box below.