v1.0.0 — Initial Public Benchmark Release

This is the initial public release of the pydre parallelism benchmark, a workload-aware study of parallel execution strategies in real-world analytics pipelines.

Highlights

• Systematic benchmark of sequential, threading, and multiprocessing execution modes
• Four workload profiles: light, medium, heavy, ROI-heavy
• Empirical evaluation of worker scaling behavior and CPU utilization
• Demonstrates diminishing returns of thread-based parallelism beyond moderate worker counts
• Introduces a dynamic worker allocation strategy (~75% of logical CPUs) as a stable, near-optimal default

Key Takeaway

Parallelism is not a free performance boost. Effective parallel execution depends on aligning the execution model with workload characteristics and overhead behavior.

Included

• Benchmark runner and analysis scripts
• Workload configuration profiles (.toml)
• Aggregated results and visualizations
• Full technical report (PDF) detailing methodology, results, and conclusions

Data Policy

Large-scale experimental input data is intentionally excluded.
The benchmark pipeline is fully reusable with user-generated or domain-specific pydre-compatible datasets.

This release establishes a reproducible baseline for evaluating parallel execution strategies in pydre-based analytics workflows.