Home » Forums » METIS » Matrix Ordering

mtmetis_nd - consistent results and robustness

Will mtmetis_nd() provide the same reordering on different numbers of threads?

And how mature is this? We are considering tying it into a commercial application currently using Metis/ParMetis and concerned about robustness.

‹ Configuration paramters for optimal ordering What is the largest ever known matrix that Metis can process and what about the efficiency? ›
Submitted by bheld on Wed, 2015-09-23 16:05
»

RE: The coarsening in mt-metis is

The coarsening in mt-metis is non-deterministic with multiple threads, and as such reorderings will differ between successive runs (even with the same number of threads and random seed).

The current version of mt-metis (0.4.4) should still be considered experimental at this point (i.e., I advise not linking a commercial application to it yet). I am currently working on improving the stability of the codebase, and will hopefully release a "mature" version later this fall. The API interface will likely to change to mimick that of Metis.

Thank you,

Dominique

Submitted by dominique on Sun, 2015-09-27 22:49.
»