In an effort to make STAR-CCM+ even faster on a broader range of problem types STAR-CCM+ v9.04 (which is released next week) will offer Concurrent Per-Part Meshing (which is a bit of a mouth full, so we'll call it CPPM instead).
Now, I'm sure you're asking yourself, ‘What’s the difference between CPPM and regular parallel meshing?’
The answer is simple: Parallel meshing will take a single Part or group of parts, split the domain into arbitrary volumes, then distribute those volumes across nodes to mesh in parallel. CPPM allows users to mesh individual parts in serial but each part being meshed will be distributed to an individual processor. Since this process is really serial at its heart that means that it applies to not just the volume meshers but also the surface wrapper and the remesher. The CPPM option is only available when the part operation is in per-part mode, since the parallel delineation is deined by the part itself.
So let’s take an example. We want to non-conformally mesh a circuit board that contains 100 components. Historically STAR-CCM+ would take each part and mesh them one after another on a single processor. Now with CPPM that process can leverage as many CPUs as the user can throw at it. If we load the above simulation on 101 cores each part will be meshed on a separate core all at the same time. So now the meshing process will only take as long as it take to mesh the largest (mesh-count) component. To put some numbers behind this, If those 100 parts that each only take 10 seconds to mesh STAR-CCM+ used to take 1000 seconds to mesh, now it will only take about 10 seconds. Notice earlier we mentioned that we need 101 cores to mesh each part on one core. That extra core is the ‘controller’ core; which hides a very powerful feature. This core acts like a head node on a queuing system, so if we have 10 cores meshing 100 parts the controller will automatically monitor each core's progress and send a new part to it when it finishes its current part. This means if a user has 99 small solid components and 1 large air domain as the air domain meshes the other parts will be churning out on the remaining cores without having to wait on the air domain.
Concurrent Per-Part Meshing
Let's finish by investigating how CPPM might be used to mesh a complex problem, such as the a thermal plume analysis on a helicopter. In this case we need: A wrapped Volume of the Air, a wrapped volume of the internal exhaust passages, and 947 solid components. In previous versions of STAR-CCM+ this example took over 7 hours to complete. Now, with concurrent meshing this process can take as little as 1 and a half hours.