how to split matrix in cluster_sparse_solver, similar number of rows or similar number of nonzero elements

[Chaowen_G_]

Hi:

   I need to use cluster_sparse_solver to solve a larger complex symmetric matrix. So I need to split matrix into two parts and put them into two different compute nodes respectively. The first part is from row 1 to row n, the second part is from row n to the last row. Since the selection of row n is arbitrary, which selection will give the best performance, tow compute nodes have similar number of rows or similar number of nonzero elements?

[Gennady_F_Intel]

from performance point of view -- splitting by the equal parts of rows -- will produce the best performance.