[Artifact] One Join Order Does Not Fit All: Reducing Intermediate Results with Per-Split Query Plans

Clone the Repo

git clone git@github.com:hyj2003/SplitJoin.git

Getting Started

Dependency

Install DuckDB for Python:

pip install duckdb==v1.3.0

Install Umbra:

docker pull umbradb/umbra:25.07.1

Prepare Datasets

All datasets are available from the links below:

• WGPB (wikidata-wcg-filtered.nt.bz2)
• Orkut
• GPlus
• USPatent
• Skitter
• Topcats

---

Run End-to-end Tests

DuckDB

Create several necessary folders first:

mkdir -p duckdb_baseline_profile
mkdir -p duckdb_split_profile/{q1,q2,q3,q4,q5,q6,q7,q8,q9,q10,q11}

To generate test results for DuckDB, please run the following commands (eta: 1d):

python3 duckdb_baseline.py > duckdb_baseline.log
python3 scripts/extract_duckdb_baseline.py

To generate test results for SplitJoin in DuckDB, please run the following commands (eta: 1d):

bash duckdb_split.sh > duckdb_split.log
python3 scripts/extract_duckdb_split.py

For intermediate sizes, please see logs in the duckdb_baseline_profile folder and the duckdb_split_profile folder.

---

Umbra

To reproduce the results for SplitJoin in Umbra, first generate SQL queries for splits using:

bash gen_split_umbra_sql.sh

Create an empty database and load the data:

docker run -it -v umbra-db:/var/db -v ./:/mnt umbradb/umbra:25.07.1 umbra-sql -createdb /var/db/umbra.db
docker run -it -v umbra-db:/var/db -v ./:/mnt umbradb/umbra:25.07.1 umbra-sql /var/db/umbra.db /mnt/umbra/load.sql

Then run the following commands (eta: 1d):

bash run_umbra_split.sh
python3 scripts/extract_umbra_split.py

To reproduce the results for Umbra baseline settings, change the UMBRA_MULTIWAY environment variable to d or e, or comment out this variable (representing enforced binary join, multi-way join, and the default setting respectively), and run:

bash run_umbra_baseline.sh
python3 scripts/extract_umbra_baseline.py

with each environment setting (eta: 1d each setting).

For intermediate sizes, add EXPLAIN ANALYZE at the beginning of each query.

---

Other Experiments

To test the overall time and maximum intermediate sizes under different thresholds, run:

python3 effectiveness_study/eval_thresholds.py > effectiveness_study/eval_thresholds.log

To test the overall time of splitting one table at a time, run:

bash effectiveness_study/split_table.sh > effectiveness_study/split_table.log

To test the overall time of arbitrarily choosing a split set, run:

bash effectiveness_study/split_arbitrarily.sh > effectiveness_study/split_arbitrarily.log