gcc 4.4.7 or 4.8.5
CUDA 5.5, 6.0, 6.5, 7.0, 7.5 (tested and works)
MPI: MPICH version 3.3 or higher, corresponding OpenMPI should also work
GPU: C2050, C2070, K20, K40 (tested)
make
Type: "./gpu_ibfs.bin" it will show you what is needed.
Atention, please make concurrent-count as 128
You could use the code from "tuple_text_to_binary_csr" folder to convert a edge list (text formated) graph into CSR files (binary), e.g., if you have a text file called "test.txt", it will convert it into "test.txt_beg_pos.bin" and "test.txt_csr.bin". You will need these two files to run gpu_ibfs.
- Compile: make
- To execute: type "./text_to_bin", it will show you what is needed
- Basically, you could download a tuple list file from snap. Afterwards, you could use this converter to convert the edge list into CSR format.
For example:
- Download https://snap.stanford.edu/data/com-Orkut.html file. unzip it.
- ./text_to_bin.bin soc-orkut.mtx 1 2(the number may change due to how many lines are not edges)
- You will get soc-orkut.mtx_beg_pos.bin and soc-orkut.mtx_csr.bin.
- You could use these two files to run enterprise.
- Centrality computation, e.g., Betweenness centrality, Closeness centrality.
- Multi-source shortest path, All-pairs shortest path.
- Reachiability index construction.
- In short, any applications, which need to run multiple traversals on the same graph, should benefit
[SC '15] Enterprise: Breadth-First Graph Traversal on GPUs [PDF] [Slides] [Blog]
[SIGMOD '16] iBFS: Concurrent Breadth-First Search on GPUs [PDF] [Slides] [Poster]
