TCUDB

Requirements

Some Python modules are require for translate.py(need to use Python2.7 to run).

config
pybind11
pandas
scipy
numpy
sklearn
os

Generate shrunken graph from SNAP-penn_road

We provide penn_road.ipynb to generate node.tbl, edge.tbl, outdegree.tbl and pagerank.tbl.

How to run

Use scripts in trunk directory to run.

./run_em_beer.sh

Manually run queries

SQL to XML.

./translate.py test/entity_match/beer.schema

Load data.

cd src/utility/
make loader
./gpuDBLoader --tablea ../../test/dbgen/beer_tableA.tbl --tableb ../../test/dbgen/beer_tableB.tbl

Compute metadata and generate CUDA driver code.

Back to trunk directory.

cd ../../
./translate.py test/entity_match/beer.sql test/entity_match/beer.schema test/dbgen/beer_tableA.tbl test/dbgen/beer_tableB.tbl

Run a query.

src/cuda/TCUDB --datadir src/utility/

Note: This experiment was conducted using NVIDIA RTX3090: -gencode arch=compute_80,code=sm_80 with Driver Version: 460.32.03 CUDA Version: 11.2

• trunk/test/dbgen/denseTblGen.py use to generate tables for msplit -- python denseTblGen.py --dim 8192
• trunk/metaGen.py precompute metadata for the estimation