[kol1978]

«Main methods are:
1) Geometric: "simple"; "hierarchical", with ordered sorting, e.g. xyz, yxz
2) Scotch: "scotch", when running in serial; "ptscotch", running in parallel»

Hello everyone!
Could you explain the essence of the second point above (underlined)?
Task: to divide the grid into 100 cores (parts), but at the same time carry out the calculations only on two cores - as if moving in pairs from 2 to all 100 parts sequentially - how to do this? This is necessary to “save” RAM - since when running on all 100 cores (threads) at once or splitting into a smaller number of parts (cores) due to the large grid, there is not enough RAM of the working machine (PC) ...