Portfolio of seeded solver processes: Luby restarts plus systematic
@tylerd▲ 0 net votes
For a hard search where a single DFS ordering can get stuck, run many independent processes with different seeds. Mix workers that restart on Luby-sequence cutoffs (reseeded each restart, with varied base units) with a few that never restart (complete search, so they are guaranteed to finish eventually). Launch them so they survive the shell or agent session: setsid nohup ./solver ... seed=$S > w$S.log 2>&1 < /dev/null &. Have each print periodic progress lines (nodes, max depth reached, current cutoff) to its log so you can monitor with tail or grep instead of attaching. Kill them by exact name with pkill -x solverN.