How simulations too expensive for one supercomputer can be done with two supercomputers

What is it about?

Plasma turbulence simulations can be incredibly expensive in terms of compute resources, because instead of "normal" three-dimensional space, they have to be solved in five- or six-dimensional space. To know what happens in detail, the necessary simulation can become too large for the memory of a single supercomputer. We have shown how such an expensive simulation can be distributed to two supercomputers: One ingredient is that the data is written as a file on each system, and then these files are sent through the internet to the other system. The other ingredient is an optimized way of distributing the simulation onto the two systems. If you are now thinking: "why don't they put the left part on one computer and the right part on the other?", this is exactly what we are not doing, because it would be too much data to exchange this way. Instead, each system considers some dimensions in far more detail, with far more resolution, than the other system. As a result, only a tiny fraction of the data in the simulation has to be communicated across the internet.

Featured Image

Photo by Joshua Sortino on Unsplash

Photo by Joshua Sortino on Unsplash

Why is it important?

Enabling plasma turbulence simulations at extremely fine scales may help better understand controlled fusion processes. Controlled fusion is a potential source of abundant, low-cost energy.

Perspectives