This repository contains a collection of experiments to test the effects of IMV. It is mainly composed of two parts:
(1) Testing the effects of IMV on individual operators, including the hash join probe, binary tree search, hash join build and hash aggregation. The source codes lay at src/imv-operator/. This part is based on the source code of the work.
(2) Comparing IMV with other three state-of-the-art query engines: Compilation, Vectorization, and ROF. The source codes lay at src/imv/. This part is based on the code base Database Engines: Vectorization vs. Compilation
A configuration file is provided to build this project with CMake. In the project directory run:
mkdir -p build/release
cd build/release
cmake -DCMAKE_BUILD_TYPE=Release -DVECTORWISE_BRANCHING=on -DKNL=OFF ../..
make
Generating two main binaries:
(1) operator:
(1.1) generating data:
./operator -a GEN -n 32 --r-size=1048576 --s-size=52428800 --r-skew=1 --s-skew=1
(1.2) testing individual operators: -a: Applications:(NPO: hash join probe, BTS: binary tree search, BUILD and AGG), -n threads number.
./operator -a NPO -n 32 --r-file=r_skew=1_size=1M --s-file=s_skew=1_size=50M_max=1M
(2) engine: 3 parameters: reteated times, data source and threads number
./engine 5 ../data/tpch/sf100 32
scripts/test.sh can be used to run experiments of the paper.
(1) reading raw data from DBGEN ends without '|', so change it in Line:146 @ Import.cpp
(2)
terminate called after throwing an instance of 'std::runtime_error' what(): Buffer 20 not found
This error results from the Buffer 20 is not created but it is used at /src/benchmarks/sbb/queries/q4X.cpp. Changes lay at those files using "#define CHANGE".