Datasets:

YixiangHu
/

L-RIPLIB

license: mit
language:
  - en
tags:
  - optimization
  - planning
pretty_name: L-RIPLIB
size_categories:
  - 1K<n<10K
configs:
  - config_name: default
    data_files:
      - split: Easy
        path: data/Easy/*.json
      - split: Normal
        path: data/Normal/*.json
      - split: Hard
        path: data/Hard/*.json

L-RIPLIB

Dataset Summary

L-RIPLIB is an industrial-scale benchmark for Resource Investment Problems (RIP) derived from cloud computing workloads. It contains 1,000 instances with problem sizes ranging from 2,500 to 10,000 tasks, intended to support realistic large-scale evaluation and to complement smaller classical benchmarks (e.g., PSPLIB).

Each instance is stored as a JSON record describing a task set with time windows, durations, precedence constraints, per-task resource requirements, and solution-related metadata produced by OR-Tools CP-SAT under a time cap.

Supported Tasks and Usage Scenarios

This dataset is suitable for:

Large-scale project/task scheduling with precedence constraints and time windows.
Resource provisioning / resource investment with per-resource unit costs.
Learning-augmented optimization (e.g., predicting good schedules, costs, bounds, or warm-start solutions).
Dynamic / continual re-optimization experiments using the provided “modified_data” deltas (see “Modified_data” field).

Languages

English

Dataset Structure

Data Format

One JSON object per instance.

Data Fields (per instance)

The dataset uses the following key elements:

Tasks (T): list of task names (activities) within the instance.
Earliest_start (e): earliest start time for each task.
Deadline (l): deadline / latest finish time for each task.
Duration (d): duration for each task.
Dependencies (P): precedence constraints specifying which tasks must finish before others can start.
Resources (R): resources allocated to each task (resource requirements).
Costs (c): unit cost of each resource type.
Task_start ((S_i)_{i∈T}): a CP-SAT solution (task start times) obtained under a limited time budget of 0.1 × |T| seconds.
Best_cost: total resource cost for the provided solution.
Time: CP-SAT solve time for the instance.
Bound: CP-SAT lower bound on total resource cost.
Modified_data (Δq): the difference between q and q' (used to represent instance modifications).

Citation

If you find our work helpful, feel free to give us a cite.

@misc{hu2026ischedulerreinforcementlearningdrivencontinual,
      title={iScheduler: Reinforcement Learning-Driven Continual Optimization for Large-Scale Resource Investment Problems}, 
      author={Yi-Xiang Hu and Yuke Wang and Feng Wu and Zirui Huang and Shuli Zeng and Xiang-Yang Li},
      year={2026},
      eprint={2602.06064},
      archivePrefix={arXiv},
      primaryClass={cs.DC},
      url={https://arxiv.org/abs/2602.06064}, 
}