Liberator: A Data Reuse Framework for Out-of-Memory Graph Computing on GPUs

Liberator is an out-of-GPU-memory graph processing framework.

Compilation

To compile the Liberator. You need have cmake, g++ and CUDA 11.4 toolkit. You should enter the project root dir. Create a directory ie. cmake-build-debug. Enter it and cmake .. and make to complile the project. You might need to change the cuda path according to your environment in CMakeList.txt

Input graph formats

Liberator accepts the binary CSR format just like :

0 4 7 9
1 2 3 4 2 3 4 3 4 5 6

Data format converter

There is a converter which can convert txt to CSR in the folder converter. .bcsr is for bfs and cc, .bcsc is for pr(pagerank) and .bwcsr is for sssp.

Running

$ ./Liberator 
--input datapath 
--type bfs (graph processing algorithm)
--sourceNode 0 (only for bfs and sssp)
--model 7 or 0 (7 is Liberator and the 0 model is our previous work Ascetic )
--tetsTime n (the algorithm will excute n times)

Publication

[IEEE Transactions on Parallel and Distributed Systems 24 April 2023 doi: DOI 10.1109/TPDS.2023.3268662] Shiyang Li, Ruiqi Tang, Jingyu Zhu, Ziyi Zhao, Xiaoli Gong, Jin Zhang, Wen-wen Wang, and Pen-Chung Yew. Liberator: A Data Reuse Framework for Out-of-Memory Graph Computing on GPUs.