@inproceedings{
  author={ H. Wang, S. Potluri, M. Luo, A. Singh, X. Ouyang, S. Sur, D. Panda },
  title={ Optimized Non-contiguous MPI Datatype Communication for GPU Clusters: Design Implementation and Evaluation with MVAPICH2 },
  conference={ IEEE Cluster '11 },
  year={ 2011 },
  month={ September },
  location={ None },
  source={ http://nowlab.cse.ohio-state.edu/publications/ },
}