GPU-based Multi-way Join Operators

Accelerating the evaluation of multi-table join queries in databases and graphs using multi-way join (AMHJ) and worst-case optimal join (ALFTJ).

Compile

Compile the dependency cub before compiling our code.

mdkir dependencies/cudpp/build
cd dependencies/cudpp/build
cmake ..
make -j16
cd ../../..

Then, compile our code by the following commands.

mkdir MHJ-GPU/build
cd MHJ-GPU/build
cmake ..
make -j16
cd ../..
mkdir LFTJ-GPU/build
cd LFTJ-GPU/build
cmake ..
make -j16
cd ../..

The versions of the software we use is listed as follows.

cmake: 3.20.2
Make: 4.2.1
GCC: 8.5.0
cuda: 10.2

Use the following command to run AMHJ.

./MHJ-GPU/build/exec-AMHJ <query-type> <dataset-path> <ooc> <ws> <dro> <fib>

The description and options of each parameter are listed in the following table.

Parameters	Description	Valid Value
query-type	Type of query	NORMAL/TRI/FOUR
dataset-path	Path of the query	N/A
ooc	Disable/Enable out of core support	0/1
ws	Disable/Enable work sharing	0/1
dro	Disable/Enable direct result output	0/1
fib	Disable/Enable frequent item buffering	0/1

Use the following command to run ALFTJ.

./LFTJ-GPU/build/exec-LFTJ <dataset-path> <algo-type> <ooc> <ws>

The description and options of each parameter are listed in the following table.

Parameters	Description	Valid Value
dataset-path	Path of the query	N/A
algo-type	Type of algorithm (BFS/DFS)	0/1
ooc	Disable/Enable out of core support	0/1
ws	Disable/Enable work sharing	0/1