A practically constant-time MPI broadcast algorithm for large-scale InfiniBand clusters with multicast. Hoefler, T., Siebert, C., & Rehm, W. In Proceedings - 21st International Parallel and Distributed Processing Symposium, IPDPS 2007; Abstracts and CD-ROM, 2007. ![A practically constant-time MPI broadcast algorithm for large-scale InfiniBand clusters with multicast [link]](https://bibbase.org/img/filetypes/link.svg "link")
Website doi abstract bibtex An efficient implementation of the MPLBCAST operation is crucial for many parallel scientific applications. The hardware multicast operation seems to be applicable to switch-based InfiniBand cluster systems. Several approaches have been implemented so far, however there has been no production-ready code available yet. This makes optimal algorithms to a subject of active research. Some problems still need to be solved in order to bridge the semantic gap between the unreliable multicast and MPLBCAST. The biggest of those problems is to ensure the reliable data transmission in a scalable way. Acknowledgement- based methods that scale logarithmically with the number of participating MPI processes exist, but they do not meet the supernormal demand of high-performance computing. We propose a new algorithm that performs the MPLBCAST operation in a practically constant time, independent of the communicator size. This method is well suited for large communicators and (especially) small messages due to its good scaling and its ability to prevent parallel process skew. We implemented our algorithm as a collective component for the Open MPI framework using native InfiniBand multicast and we show its scalability on a cluster with 1.1.6 compute nodes, where it saves up to 41% MPLBCAST latency in comparison to the "TUNED" Open MPI collective. © 2007 IEEE.
@inproceedings{
title = {A practically constant-time MPI broadcast algorithm for large-scale InfiniBand clusters with multicast},
type = {inproceedings},
year = {2007},
keywords = {Algorithms,Codes (symbols); Data communication equipment; Mul,Hardware multicast operation; InfiniBand cluster},
websites = {https://www.scopus.com/inward/record.uri?eid=2-s2.0-34548793392&doi=10.1109%2FIPDPS.2007.370475&partnerID=40&md5=39cc41b3df0b0710dd8d9063886de7c1},
city = {Long Beach, CA},
id = {f5d59355-0222-3790-b84e-fbd4ac5ba24f},
created = {2018-01-09T20:30:38.040Z},
file_attached = {false},
profile_id = {42d295c0-0737-38d6-8b43-508cab6ea85d},
last_modified = {2018-03-12T19:03:18.613Z},
read = {false},
starred = {false},
authored = {true},
confirmed = {true},
hidden = {false},
citation_key = {Hoefler2007},
source_type = {conference},
notes = {cited By 26; Conference of 21st International Parallel and Distributed Processing Symposium, IPDPS 2007 ; Conference Date: 26 March 2007 Through 30 March 2007; Conference Code:70236},
folder_uuids = {2aba6c14-9027-4f47-8627-0902e1e2342b},
private_publication = {false},
abstract = {An efficient implementation of the MPLBCAST operation is crucial for many parallel scientific applications. The hardware multicast operation seems to be applicable to switch-based InfiniBand cluster systems. Several approaches have been implemented so far, however there has been no production-ready code available yet. This makes optimal algorithms to a subject of active research. Some problems still need to be solved in order to bridge the semantic gap between the unreliable multicast and MPLBCAST. The biggest of those problems is to ensure the reliable data transmission in a scalable way. Acknowledgement- based methods that scale logarithmically with the number of participating MPI processes exist, but they do not meet the supernormal demand of high-performance computing. We propose a new algorithm that performs the MPLBCAST operation in a practically constant time, independent of the communicator size. This method is well suited for large communicators and (especially) small messages due to its good scaling and its ability to prevent parallel process skew. We implemented our algorithm as a collective component for the Open MPI framework using native InfiniBand multicast and we show its scalability on a cluster with 1.1.6 compute nodes, where it saves up to 41% MPLBCAST latency in comparison to the "TUNED" Open MPI collective. © 2007 IEEE.},
bibtype = {inproceedings},
author = {Hoefler, T and Siebert, C and Rehm, W},
doi = {10.1109/IPDPS.2007.370475},
booktitle = {Proceedings - 21st International Parallel and Distributed Processing Symposium, IPDPS 2007; Abstracts and CD-ROM}
}
Downloads: 0
{"_id":"D4GZwvheetLNb7HzZ","bibbaseid":"hoefler-siebert-rehm-apracticallyconstanttimempibroadcastalgorithmforlargescaleinfinibandclusterswithmulticast-2007","downloads":0,"creationDate":"2018-03-12T19:10:28.089Z","title":"A practically constant-time MPI broadcast algorithm for large-scale InfiniBand clusters with multicast","author_short":["Hoefler, T.","Siebert, C.","Rehm, W."],"year":2007,"bibtype":"inproceedings","biburl":"https://bibbase.org/service/mendeley/42d295c0-0737-38d6-8b43-508cab6ea85d","bibdata":{"title":"A practically constant-time MPI broadcast algorithm for large-scale InfiniBand clusters with multicast","type":"inproceedings","year":"2007","keywords":"Algorithms,Codes (symbols); Data communication equipment; Mul,Hardware multicast operation; InfiniBand cluster","websites":"https://www.scopus.com/inward/record.uri?eid=2-s2.0-34548793392&doi=10.1109%2FIPDPS.2007.370475&partnerID=40&md5=39cc41b3df0b0710dd8d9063886de7c1","city":"Long Beach, CA","id":"f5d59355-0222-3790-b84e-fbd4ac5ba24f","created":"2018-01-09T20:30:38.040Z","file_attached":false,"profile_id":"42d295c0-0737-38d6-8b43-508cab6ea85d","last_modified":"2018-03-12T19:03:18.613Z","read":false,"starred":false,"authored":"true","confirmed":"true","hidden":false,"citation_key":"Hoefler2007","source_type":"conference","notes":"cited By 26; Conference of 21st International Parallel and Distributed Processing Symposium, IPDPS 2007 ; Conference Date: 26 March 2007 Through 30 March 2007; Conference Code:70236","folder_uuids":"2aba6c14-9027-4f47-8627-0902e1e2342b","private_publication":false,"abstract":"An efficient implementation of the MPLBCAST operation is crucial for many parallel scientific applications. The hardware multicast operation seems to be applicable to switch-based InfiniBand cluster systems. Several approaches have been implemented so far, however there has been no production-ready code available yet. This makes optimal algorithms to a subject of active research. Some problems still need to be solved in order to bridge the semantic gap between the unreliable multicast and MPLBCAST. The biggest of those problems is to ensure the reliable data transmission in a scalable way. Acknowledgement- based methods that scale logarithmically with the number of participating MPI processes exist, but they do not meet the supernormal demand of high-performance computing. We propose a new algorithm that performs the MPLBCAST operation in a practically constant time, independent of the communicator size. This method is well suited for large communicators and (especially) small messages due to its good scaling and its ability to prevent parallel process skew. We implemented our algorithm as a collective component for the Open MPI framework using native InfiniBand multicast and we show its scalability on a cluster with 1.1.6 compute nodes, where it saves up to 41% MPLBCAST latency in comparison to the \"TUNED\" Open MPI collective. © 2007 IEEE.","bibtype":"inproceedings","author":"Hoefler, T and Siebert, C and Rehm, W","doi":"10.1109/IPDPS.2007.370475","booktitle":"Proceedings - 21st International Parallel and Distributed Processing Symposium, IPDPS 2007; Abstracts and CD-ROM","bibtex":"@inproceedings{\n title = {A practically constant-time MPI broadcast algorithm for large-scale InfiniBand clusters with multicast},\n type = {inproceedings},\n year = {2007},\n keywords = {Algorithms,Codes (symbols); Data communication equipment; Mul,Hardware multicast operation; InfiniBand cluster},\n websites = {https://www.scopus.com/inward/record.uri?eid=2-s2.0-34548793392&doi=10.1109%2FIPDPS.2007.370475&partnerID=40&md5=39cc41b3df0b0710dd8d9063886de7c1},\n city = {Long Beach, CA},\n id = {f5d59355-0222-3790-b84e-fbd4ac5ba24f},\n created = {2018-01-09T20:30:38.040Z},\n file_attached = {false},\n profile_id = {42d295c0-0737-38d6-8b43-508cab6ea85d},\n last_modified = {2018-03-12T19:03:18.613Z},\n read = {false},\n starred = {false},\n authored = {true},\n confirmed = {true},\n hidden = {false},\n citation_key = {Hoefler2007},\n source_type = {conference},\n notes = {cited By 26; Conference of 21st International Parallel and Distributed Processing Symposium, IPDPS 2007 ; Conference Date: 26 March 2007 Through 30 March 2007; Conference Code:70236},\n folder_uuids = {2aba6c14-9027-4f47-8627-0902e1e2342b},\n private_publication = {false},\n abstract = {An efficient implementation of the MPLBCAST operation is crucial for many parallel scientific applications. The hardware multicast operation seems to be applicable to switch-based InfiniBand cluster systems. Several approaches have been implemented so far, however there has been no production-ready code available yet. This makes optimal algorithms to a subject of active research. Some problems still need to be solved in order to bridge the semantic gap between the unreliable multicast and MPLBCAST. The biggest of those problems is to ensure the reliable data transmission in a scalable way. Acknowledgement- based methods that scale logarithmically with the number of participating MPI processes exist, but they do not meet the supernormal demand of high-performance computing. We propose a new algorithm that performs the MPLBCAST operation in a practically constant time, independent of the communicator size. This method is well suited for large communicators and (especially) small messages due to its good scaling and its ability to prevent parallel process skew. We implemented our algorithm as a collective component for the Open MPI framework using native InfiniBand multicast and we show its scalability on a cluster with 1.1.6 compute nodes, where it saves up to 41% MPLBCAST latency in comparison to the \"TUNED\" Open MPI collective. © 2007 IEEE.},\n bibtype = {inproceedings},\n author = {Hoefler, T and Siebert, C and Rehm, W},\n doi = {10.1109/IPDPS.2007.370475},\n booktitle = {Proceedings - 21st International Parallel and Distributed Processing Symposium, IPDPS 2007; Abstracts and CD-ROM}\n}","author_short":["Hoefler, T.","Siebert, C.","Rehm, W."],"urls":{"Website":"https://www.scopus.com/inward/record.uri?eid=2-s2.0-34548793392&doi=10.1109%2FIPDPS.2007.370475&partnerID=40&md5=39cc41b3df0b0710dd8d9063886de7c1"},"biburl":"https://bibbase.org/service/mendeley/42d295c0-0737-38d6-8b43-508cab6ea85d","bibbaseid":"hoefler-siebert-rehm-apracticallyconstanttimempibroadcastalgorithmforlargescaleinfinibandclusterswithmulticast-2007","role":"author","keyword":["Algorithms","Codes (symbols); Data communication equipment; Mul","Hardware multicast operation; InfiniBand cluster"],"metadata":{"authorlinks":{}},"downloads":0},"search_terms":["practically","constant","time","mpi","broadcast","algorithm","large","scale","infiniband","clusters","multicast","hoefler","siebert","rehm"],"keywords":["algorithms","codes (symbols); data communication equipment; mul","hardware multicast operation; infiniband cluster"],"authorIDs":[],"dataSources":["zgahneP4uAjKbudrQ","ya2CyA73rpZseyrZ8","2252seNhipfTmjEBQ"]}