MCTS Performs Worse Than Randomized-Restart List Scheduling
October 12, 2026
A controlled study of DAG workflow scheduling finds that Monte Carlo Tree Search (MCTS) is 2.88% less efficient than a simpler randomized-restart list scheduler when given an equal computational budget.
HOW THIS AFFECTS YOU
●
builderConsider simpler randomized-restart methods for workflow scheduling to achieve better efficiency for the same compute cost.
●
researcherThis challenges the assumption that MCTS is inherently superior for DAG scheduling tasks.