Abstract
Parallel computations are intrinsically non-reproducible, due to a combined effect of non-deterministic parallel reductions and non-associative floating point operations. Different strategies have been proposed in literature to alleviate this issue or eliminate it altogether, however at present there is no study on the performance impact of associative floating point operations on large scale applications. In this work, we implement associative operations using binned doubles in MiniFE, and perform various performance tests on Cirrus and Fulhame, two state-of-the-art HPC systems.
This is a preview of subscription content, log in via an institution to check access.
Access this chapter
Tax calculation will be finalised at checkout
Purchases are for personal use only
References
Cirrus UK National Tier-2 HPC service (2019) http://www.cirrus.ac.uk
Ahrens, P., Nguyen, H.D., Demmel, J.: Efficient reproducible floating point summation and BLAS. Technical report UCB/EECS-2015-229, EECS Department, University of California, Berkeley, December 2015
Heroux, M.A., et al.: Improving performance via mini-applications. Technical report SAND2009-5574, Sandia National Laboratories (2009)
Kahan, W.: Pracniques: further remarks on reducing truncation errors. Commun. ACM 8(1), 40 (1965)
OpenMP Architecture Review Board: OpenMP application program interface version 3.1 (2001). https://www.openmp.org/wp-content/uploads/OpenMP3.1.pdf
Robey, R.W., Robey, J.M., Aulwes, R.: In search of numerical consistency in parallel programming. Parallel Comput. 37(4–5), 217–229 (2011)
Vandevoorde, D., Josuttis, N.M.: C++ Templates. Addison-Wesley Longman Publishing Co., Inc., Boston (2002)
Acknowledgement
This research is supported by Rolls-Royce plc through the EPSRC ASiMoV Prosperity Partnership Project. The authors would like to thank Rolls-Royce plc for granting permission to publish this work. The experiments were undertaken on two different systems: the Cirrus UK National Tier-2 HPC Service at EPCC, funded by the University of Edinburgh and EPSRC (EP/P020267/1); and the Fulhame system, which is supplied to EPCC as part of the Catalyst UK program, a collaboration with Hewlett Packard Enterprise, Arm and SUSE to accelerate the adoption of Arm based supercomputer applications in the UK.
Author information
Authors and Affiliations
Corresponding author
Editor information
Editors and Affiliations
Rights and permissions
Copyright information
© 2019 Springer Nature Switzerland AG
About this paper
Cite this paper
Bombace, N., Weiland, M. (2019). A Study on the Performance of Reproducible Computations. In: Weiland, M., Juckeland, G., Alam, S., Jagode, H. (eds) High Performance Computing. ISC High Performance 2019. Lecture Notes in Computer Science(), vol 11887. Springer, Cham. https://doi.org/10.1007/978-3-030-34356-9_33
Download citation
DOI: https://doi.org/10.1007/978-3-030-34356-9_33
Published:
Publisher Name: Springer, Cham
Print ISBN: 978-3-030-34355-2
Online ISBN: 978-3-030-34356-9
eBook Packages: Computer ScienceComputer Science (R0)