graphlet-counting

Graphlet counting implementation in CUDA

Running on CRC cluster.

Not sure how much larger the graphs will get, but, there is interactive access to the 2 main CRC GPU machines. If the jobs don't run long, then this will probably be enough.

General procedure for compiling and running (without submitting to the queue).

Log into one of the CRC frontend machines: netid@crc*.crc.nd.edu (crcfe01.crc.nd.edu, crcfe02.crc.nd.edu, crcfeIB01.crc.nd.edu)

To get the interactive shell, run the following command for the NVIDIA GeForce GTX 1080 GPU

[snjoroge@crcfe01 ~]$ qrsh -q gpu-debug

or, for the NVIDIA GeForce GTX TITAN X GPU

[snjoroge@crcfe01 ~]$ qrsh -q gpu

I don't think you can log into the GPU machines directly. I tried and failed. Prompt will change to the machine you are on.

Then, load the cuda compiler (Default is version 7.5)

[snjoroge@qa-titanx-001]$ module load cuda

Locate your code in your AFS space. Compile with the follow command.

[snjoroge@qa-titanx-001 graphlet-counting]$ nvcc -std=c++11 -o graphlests graphlets.cu

Run the program with a graph

[snjoroge@qa-titanx-001 graphlet-counting]$ ./graphlets data/network.txt 256