DEBS 2022 Case Study

[GnaneshGnani]

Summary

DEBS 2022 Grand Challenge evaluation prep: documentation, analysis tooling, and a script to fetch the public CSV corpus. Large CSVs under data/ and data_filtered/ are not committed (see datasets_eval/debs/.gitignore); use the script and filter_data.py locally.

Dataset

• Corpus: Zenodo 6382482
• Paper: arXiv:2206.13237 — query definitions and challenge description
• Streams: full feed (data/) vs last-trade filtered (data_filtered/); methodology in DEBS_2022/03_benchmark_methodology.md

What this PR adds

Path	Role
datasets_eval/debs/DEBS_2022/01_dataset_statistics.md	Field meanings, window stats, cardinality, per-query matrix
datasets_eval/debs/DEBS_2022/02_benchmark_queries.md	Q1–Q12 specs, OTLP mapping, processor hints, validation
datasets_eval/debs/DEBS_2022/03_benchmark_methodology.md	Throughput / latency / sketch-finance procedures, controller API
datasets_eval/debs/DEBS_2022/README.md	Doc index and pipeline overview
datasets_eval/debs/analysis/	Scripts and summarized results where tracked
datasets_eval/debs/download_data.sh	Downloads debs2022-gc-trading-day-*.csv into datasets_eval/debs/data/ (idempotent)

After download, build filtered CSVs:

cd datasets_eval/debs/analysis/code && python3 filter_data.py

Follow-up (separate stacked PRs)

• Controller: sketch selection and agent-generated collector YAML for DEBS workloads
• datasets_eval/debs/benchmark/: replay, Prometheus scrape, compare, per-query ground truth for Q1–Q8; Q9–Q12 are throughput/latency only (no sketch-finance GT in this stack)

Related

• Open PRs in this stack: controller → benchmark core → debs_q1 … debs_q8