The calculation time is multiplied by ten when the dimensions are divided by three. Is it a computational problem or is it a physics problem? What should I fix on Comsol to get an equivalent computing time?
I'm afraid that we will need more info to answer you properly. Which model flow are you solving for? (Stokes,laminar or turbulent). Are you using a segregated solver or a fully coupled one? In any case, can you upload a screen capture of the convergency plots? My best guess is that by scaling the size your flow behavior has changed, and probably it has smaller eddies now, making it more difficult to converge in the segregated step. If you check the number of iterations in the convergency plot you will notice this because it will take many more steps that in the bigger model. The best way to deal with that is refining the mesh to adapt it to the new gradients. Alternatively you can toggle the inconsistent stabilization option in the model physics, however that introduces an overdiffusive behaviour in the solution. Check this for more info: https://www.comsol.com/blogs/understanding-stabilization-methods/
I'm afraid that we will need more info to answer you properly. Which model flow are you solving for? (Stokes,laminar or turbulent). Are you using a segregated solver or a fully coupled one? In any case, can you upload a screen capture of the convergency plots? My best guess is that by scaling the size your flow behavior has changed, and probably it has smaller eddies now, making it more difficult to converge in the segregated step. If you check the number of iterations in the convergency plot you will notice this because it will take many more steps that in the bigger model. The best way to deal with that is refining the mesh to adapt it to the new gradients. Alternatively you can toggle the inconsistent stabilization option in the model physics, however that introduces an overdiffusive behaviour in the solution. Check this for more info: https://www.comsol.com/blogs/understanding-stabilization-methods/