GPU peer-to-peer techniques applied to a cluster interconnect
Istituto Nazionale di Fisica Nucleare, Sezione Roma Tor Vergata, Rome, Italy
arXiv:1307.8276 [physics.comp-ph], (31 Jul 2013)
@article{2013arXiv1307.8276A,
author={Ammendola}, R. and {Bernaschi}, M. and {Biagioni}, A. and {Bisson}, M. and {Fatica}, M. and {Frezza}, O. and {Lo Cicero}, F. and {Lonardo}, A. and {Mastrostefano}, E. and {Stanislao Paolucci}, P. and {Rossetti}, D. and {Simula}, F. and {Tosoratto}, L. and {Vicini}, P.},
title={"{GPU peer-to-peer techniques applied to a cluster interconnect}"},
journal={ArXiv e-prints},
archivePrefix={"arXiv"},
eprint={1307.8276},
primaryClass={"physics.comp-ph"},
keywords={Physics – Computational Physics, Computer Science – Distributed, Parallel, and Cluster Computing},
year={2013},
month={jul},
adsurl={http://adsabs.harvard.edu/abs/2013arXiv1307.8276A},
adsnote={Provided by the SAO/NASA Astrophysics Data System}
}
Modern GPUs support special protocols to exchange data directly across the PCI Express bus. While these protocols could be used to reduce GPU data transmission times, basically by avoiding staging to host memory, they require specific hardware features which are not available on current generation network adapters. In this paper we describe the architectural modifications required to implement peer-to-peer access to NVIDIA Fermi- and Kepler-class GPUs on an FPGA-based cluster interconnect. Besides, the current software implementation, which integrates this feature by minimally extending the RDMA programming model, is discussed, as well as some issues raised while employing it in a higher level API like MPI. Finally, the current limits of the technique are studied by analyzing the performance improvements on low-level benchmarks and on two GPU-accelerated applications, showing when and how they seem to benefit from the GPU peer-to-peer method.
August 1, 2013 by hgpu
