177 	 * 	ncpu(X) is CPU count of node X  in alloc_nodes_groups()
181 	 * 	ncpu(A) <= ncpu(B)  in alloc_nodes_groups()
182 	 * 	ncpu(A) + ncpu(B) = N  in alloc_nodes_groups()
185 	 * 	grps(A) = max(1, round_down(G * ncpu(A) / N))  in alloc_nodes_groups()
191 	 * 2) obviously grps(A) <= ncpu(A) because:  in alloc_nodes_groups()
193 	 * 	if grps(A) is 1, then grps(A) <= ncpu(A) given  in alloc_nodes_groups()
194 	 * 	ncpu(A) >= 1  in alloc_nodes_groups()
197 	 * 		grps(A) <= G * ncpu(A) / N <= ncpu(A), given G <= N  in alloc_nodes_groups()
199 	 * 3) prove how grps(B) <= ncpu(B):  in alloc_nodes_groups()
201 	 * 	if round_down(G * ncpu(A) / N) == 0, vecs(B) won't be  in alloc_nodes_groups()
202 	 * 	over-allocated, so grps(B) <= ncpu(B),  in alloc_nodes_groups()
207 	 * 		round_down(G * ncpu(A) / N) =  in alloc_nodes_groups()
208 	 * 		round_down((N - delta) * ncpu(A) / N) =  in alloc_nodes_groups()
209 	 * 		round_down((N * ncpu(A) - delta * ncpu(A)) / N)	 >=  in alloc_nodes_groups()
210 	 * 		round_down((N * ncpu(A) - delta * N) / N)	 =  in alloc_nodes_groups()
215 	 * 	grps(A) - G >= ncpu(A) - delta - G  in alloc_nodes_groups()
217 	 * 	G - grps(A) <= G + delta - ncpu(A)  in alloc_nodes_groups()
219 	 * 	grps(B) <= N - ncpu(A)  in alloc_nodes_groups()
226 	 * finally for each node X: grps(X) <= ncpu(X).  in alloc_nodes_groups()