Abstract
High-performance computing applications heavily rely on message-passing mechanisms for data sharing in cluster environments, and the MPI library stands as the default communication library for parallel applications. Significant efforts have been directed toward optimizing data distribution and buffering based on size. This optimization aims to enhance communication performance and prevent issues such as running out of memory on the target node.
Furthermore, the emergence of multicore clusters with larger node sizes has stimulated the investigation of hierarchical collective algorithms that consider the placement of processes within the cluster and the memory hierarchy.
This paper studies and compares the performance of the algorithm of the reduction collective from the literature, specifically several implementations that do not form part of the current MPI standard, which tackle this issue. We implement the algorithms on top of Intel MPI and OpenMPI libraries using the MPI profiling interface.
Experimental results with the Intel MPI Benchmarks on a multicore cluster, Intel Platinum processor-based and OmniPath interconnection network show much room for improvement in the performance of collectives depending on the message sizes.
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
Similar content being viewed by others
References
Bayatpour, M., Maqbool Hashmi, J., Chakraborty, S., Subramoni, H., Kousha, P., Panda, D.K.: Salar: scalable and adaptive designs for large message reduction collectives. In: 2018 IEEE International Conference on Cluster Computing (CLUSTER), pp. 12–23 (2018). https://doi.org/10.1109/CLUSTER.2018.00014
Bayatpour, M., Chakraborty, S., Subramoni, H., Lu, X., Panda, D.K.D.: Scalable reduction collectives with data partitioning-based multi-leader design. In: Proceedings of the International Conference for High Performance Computing, Networking, Storage and Analysis, SC 2017, Association for Computing Machinery, New York (2017). https://doi.org/10.1145/3126908.3126954, https://doi.org/10.1145/3126908.3126954
Castelló, A., Catalán, M., Dolz, M.F., Quintana-Ortí, E.S., Duato, J.: Analyzing the impact of the MPI allreduce in distributed training of convolutional neural networks. Computing , 1–19 (2021). https://doi.org/10.1007/s00607-021-01029-2
Chunduri, S., Parker, S., Balaji, P., Harms, K., Kumaran, K.: Characterization of mpi usage on a production supercomputer. In: SC18: International Conference for High Performance Computing, Networking, Storage and Analysis, pp. 386–400 (Nov 2018). https://doi.org/10.1109/SC.2018.00033
Giordano, A., De Rango, A., Rongo, R., D’Ambrosio, D., Spataro, W.: Dynamic load balancing in parallel execution of cellular automata. IEEE Trans. Parallel Distrib. Syst. 32(2), 470–484 (2021). https://doi.org/10.1109/TPDS.2020.3025102
Graham, R.L., Shipman, G.: MPI support for multi-core architectures: optimized shared memory collectives. In: Lastovetsky, A., Kechadi, T., Dongarra, J. (eds.) EuroPVM/MPI 2008. LNCS, vol. 5205, pp. 130–140. Springer, Heidelberg (2008). https://doi.org/10.1007/978-3-540-87475-1_21
Heroux, M.A., et al.: Improving Performance via Mini-applications. Tech. Rep. SAND2009-5574, Sandia National Laboratories (2009)
Intel MPI library: Intel MPI Library. https://www.intel.com/content/www/us/en/developer/tools/oneapi/mpi-library.html
OpenMPI Open Source HPC: OpenMPI Open Source HPC. http://www.open-mpi.org/
Rico-Gallego, J., Díaz-Martin, J.: Improving the performance of the mpi_allreduce collective operation through rank renaming. In: In Proceedings of First International Workshop on Sustainable Ultrascale Computing Systems (August 2014). https://core.ac.uk/download/pdf/30277086.pdf
Träff, J.L., Hunold, S.: Decomposing mpi collectives for exploiting multi-lane communication. In: 2020 IEEE International Conference on Cluster Computing (CLUSTER), pp. 270–280 (2020). https://doi.org/10.1109/CLUSTER49012.2020.00037
Utrera, G., Gil, M., Martorell, X.: Analyzing the performance of hierarchical collective algorithms on arm-based multicore clusters. In: 2022 30th Euromicro International Conference on Parallel, Distributed and Network-based Processing (PDP), pp. 230–233 (2022). https://doi.org/10.1109/PDP55904.2022.00043
Acknowledgements
The authors acknowledge the support of the Spanish Ministry of Education (PID2019-107255GB-C22) and the Generalitat de Catalunya (2021-SGR-01007).
Author information
Authors and Affiliations
Corresponding author
Editor information
Editors and Affiliations
Rights and permissions
Copyright information
© 2025 The Author(s), under exclusive license to Springer Nature Switzerland AG
About this paper
Cite this paper
Utrera, G., Gil, M., Martorell, X., Spataro, W., Giordano, A. (2025). Exploring Hierarchical MPI Reduction Collective Algorithms Targeted to Multicore Node Clusters. In: Sergeyev, Y.D., Kvasov, D.E., Astorino, A. (eds) Numerical Computations: Theory and Algorithms. NUMTA 2023. Lecture Notes in Computer Science, vol 14478. Springer, Cham. https://doi.org/10.1007/978-3-031-81247-7_35
Download citation
DOI: https://doi.org/10.1007/978-3-031-81247-7_35
Published:
Publisher Name: Springer, Cham
Print ISBN: 978-3-031-81246-0
Online ISBN: 978-3-031-81247-7
eBook Packages: Computer ScienceComputer Science (R0)