Study Shows Randomized-Restart List Scheduling Outperforms MCTS for DAG Workflows Under Equal Budget

Summary:A study published on arXiv compares Monte Carlo Tree Search (MCTS) and randomized-restart list scheduling methods for scheduling Directed Acyclic Graph (DAG) workflows under a controlled equal-budget protocol. The results demonstrate that the randomized-restart approach outperforms MCTS by 2.88% in terms of performance. The research indicates that MCTS's disadvantage is not due to insufficient search budget but rather its selection rule, which deterministically locks onto the first-expanded chil

---

Background and Motivation

Monte Carlo Tree Search (MCTS) is frequently proposed for scheduling Directed Acyclic Graph (DAG) workflows, but most studies fail to compare it with simpler methods under equal conditions. This research aims to evaluate the performance of MCTS against randomized-restart list scheduling under a controlled equal-budget protocol.

Experimental Design and Methodology

The study employs a controlled protocol where all scheduling methods are evaluated with identical computational budgets, i.e., the same number of candidate schedules built and evaluated. The experiments include:

• MCTS Scheduling Method: The primary subject of study.
• Randomized-Restart List Scheduling Method: The control group, which does not use a search tree.
• Textbook-UCT MCTS: A supplementary subject for comparison.

The analysis plan was locked before data collection, and an independent checker confirmed the validity of all 768 confirmation-stage schedules with correct finish times.

Key Findings

1. Performance Comparison: At a budget of 2048 evaluated schedules, MCTS is 2.88% worse than the restarts (95% CI [-3.62%, -2.15%]).
2. Budget Not a Bottleneck: In a separate exploration pool, restarts self-improve by 1.85% while MCTS gains only 0.24%. Additionally, 700 out of 720 of MCTS's first results are merely the warm start it was handed.
3. Impact of Selection Rule: The MCTS selection rule locks onto the first-expanded child, keeping its trials correlated, whereas restarts draw gains from the tails.
4. Structural Analysis: Restarts win on sparse irregular workflows, no difference is detectable on regular layered ones, and dense ones leave no room for improvement.
5. UCT Adjustment: Replacing the selection rule with textbook UCT reduces the deficit to -0.26% (95% CI [-0.71%, +0.23%]) without reversing the sign.

Engineering Trade-offs and Discussion

• Performance vs. Complexity: The randomized-restart method has lower computational complexity and is more suitable for resource-constrained environments.
• Applicable Scenarios: MCTS performs adequately in regular layered workflows but shows significant disadvantages in complex, irregular ones.
• Improvement Directions: Adjusting the selection rule or combining other optimization techniques may enhance MCTS performance.

Conclusion

This study, through controlled experiments, reveals the superiority of the randomized-restart list scheduling method for DAG workflow scheduling and provides new directions for future research.

---

Source: ArXiv AI (cs.AI) (2026-10-12)

— END —

Tags: #MCTS Follow #DAG Follow #Scheduling Algorithm Follow #Random Restart Follow #Monte Carlo Tree Search Follow