-
Notifications
You must be signed in to change notification settings - Fork 126
Main merge release/26.02 5 #836
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Main merge release/26.02 5 #836
Conversation
This PR implements the schedule-worker model for B&B, where one thread (the scheduler) is responsible for scheduling tasks to the other threads (the workers). With this model, the CPU resources can be shared efficiently among the different tasks within the MIP solver. It also allows the scheduling policy to be changed in runtime (e.g., in the future, the solver can dynamically set the number of workers allocated to each diving heuristics at runtime). This also implements a parallel reliability branching (Section 5.7 from [1]). This also fixes the incorrect pseudocost update in each node in the B&B tree. Closes NVIDIA#526. Closes NVIDIA#445. Closes NVIDIA#700. Performance over the MIPLIB2017 dataset (GH200): ``` ================================================================================ main - 046eafd (1) vs reliability-branching-cuts (2) ================================================================================ Feasible solutions = 225 vs 224 (-1) Optimal solutions = 44 vs 68 (+24) Solutions with <0.1% primal gap = 109 vs 125 (+16) Average nodes explored = 7298973 vs 4300451 (-2998522, -41.081%) Shifted geomean for MIP gap = 0.2979 vs 0.2374 (-0.0605, -20.302%) Average primal gap = 11.9935 vs 11.0833 (-0.9102, -7.589%) Average primal integral per time = 19.3326 vs 30.7054 (+11.3728, +37.038%) ================================================================================ ``` ## Reference [1] T. Achterberg, “Constraint Integer Programming,” PhD, Technischen Universität Berlin, Berlin, 2007. doi: [10.14279/depositonce-1634](https://doi.org/10.14279/depositonce-1634). Authors: - Nicolas L. Guidotti (https://github.com/nguidotti) - Chris Maes (https://github.com/chris-maes) Approvers: - Chris Maes (https://github.com/chris-maes) URL: NVIDIA#766
|
Caution Review failedThe pull request is closed. 📝 WalkthroughWalkthroughThis PR introduces a scheduler-worker paradigm for branch-and-bound optimization, adds reliability branching parameter configuration, refactors diving/best-first threading logic, enhances pseudo-cost calculations, and updates utility classes for thread safety. Key changes span solver settings, worker pool management, pseudo-cost strategies, and configuration simplification in related solvers. Changes
Estimated code review effort🎯 4 (Complex) | ⏱️ ~65 minutes Possibly related PRs
Suggested labels
Suggested reviewers
✨ Finishing touches
🧪 Generate unit tests (beta)
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
Description
Issue
Checklist
Summary by CodeRabbit
New Features
reliability_branchingparameter to control reliability-based variable selection strategy in MIP solver.Improvements
Refactor