Supplemental artifact for the ICDE submission Vivace: Exact Temporal OLAP over Interval Histories via Independent Serverless Execution.
Vivace executes exact temporal operator chains (predicate windows, duration-weighted aggregates, boolean compositions, count timelines, aligned rankings) over interval histories with stateless AWS Lambda fan-out, reading interval-clipped partition (ICP) layouts from object storage. This artifact contains the implementation, the infrastructure definitions, the benchmark configurations, the recorded measurements behind every figure, and scripts to reproduce the results at three levels of effort.
Start here: docs/artifact_evaluation_guide.md.
A one-page summary is at docs/artifact_description.pdf.
| Mode | What it shows | Needs |
|---|---|---|
Figure/table regeneration (scripts/reproduce_figures.sh) |
Every paper figure rebuilt from recorded measurements; 55 inline numbers verified | uv only, minutes |
Local correctness (scripts/run_local_correctness.sh, scripts/run_mobilitydb_generality.sh) |
Ten-query SCD2-vs-ICP exactness oracle on the real corpora; MobilityDB-TPCDS operator generality | data package (~5 GB) + public-source staging scripts |
Cloud rerun (iac/ + scripts/run_cloud_matrix.sh) |
End-to-end serverless latency/cost measurements on your own AWS account | AWS account, Terraform |
.
├── README.md
├── LICENSE / THIRD_PARTY_NOTICES.md MIT; bundled third-party components
├── pyproject.toml / uv.lock pinned Python environment (uv)
├── docs/
│ ├── artifact_evaluation_guide.md
│ ├── artifact_description.pdf one-page summary
│ ├── cost_model.md cost accounting rules
│ ├── dataset_sources.md sources, rebuild paths, licenses
│ └── query_suite.md queries, analysis questions, per-dataset predicates
├── src/
│ ├── vivace_olap/ the system (core, IR, carriers, planner,
│ │ lambda_exec async-DAG runtime, campaign,
│ │ workloads, packaging, tests)
│ └── serverless_sql_baselines/ Athena / BigQuery baseline runners
├── iac/ self-contained Terraform root module
│ ├── ec2_clickhouse/ always-on ClickHouse baseline (isolated)
│ └── ec2_mobilitydb/ always-on MobilityDB baseline (isolated)
├── experiments/
│ ├── benchmark_campaign/ campaign config (10-query final matrix)
│ ├── eval_figures/ window-scaling experiment configs + always-on drivers
│ └── external_temporal_datasets/ CAISO / Wikipedia / MobilityDB tooling
├── datasets/benchmark/ dataset bundle manifests (hashes, builds)
├── figures/ render + verification scripts
├── results/
│ ├── recorded/ recorded paper-facing measurements
│ └── figures|runs|benchmark_campaign/ your reproduction outputs (gitignored)
├── data/
│ ├── manifests/checksums.sha256 data-package integrity manifest
│ └── local/ staged datasets (gitignored; see guide)
├── scripts/ entry points for all three modes
└── tools/ ICP materialization + equivalence checks
uv run pytest src/vivace_olap/tests: 281 passed (4 skips gate on research corpora that are not part of the data package).- All reproduced figure PDFs match the manuscript figures
(
figures/VERIFICATION.md); 55/55 inline evaluation numbers verified. - ICP layouts re-derived from the SCD2 sources match the measured layouts (SpotLake 11/12 months row-identical, final live month canonically identical; CAISO 12/12 atoms row-identical).
Vivace is the paper name of the system. The implementation package is
vivace_olap; deployed AWS resources are prefixed vivace-<environment>-.
Recorded tables label the serverless runtime series VIVACE ICP async,
VIVACE SCD2 async, and VIVACE ICP cold.