Optimizing non-blocking collective operations for InfiniBand. Hoefler, T. & Lumsdaine, A. In IPDPS Miami 2008 - Proceedings of the 22nd IEEE International Parallel and Distributed Processing Symposium, Program and CD-ROM, 2008. ![Optimizing non-blocking collective operations for InfiniBand [link]](https://bibbase.org/img/filetypes/link.svg "link")
Website doi abstract bibtex Non-blocking collective operations have recently been shown to be a promising complementary approach for overlapping communication and computation in parallel applications. However, in order to maximize the performance and usability of these operations it is important that they progress concurrently with the application without introducing CPU overhead and without requiring explicit user intervention. While studying non-blocking collective operations in the context of our portable library (libNBC), we found that most MPI implementations do not sufficienctly support overlap over the InfiniBand network. To address this issue, we developed a low-level communication layer for libNBC based on the Open Fabrics InfiniBand verbs API. With this layer we are able to achieve high degrees of overlap without the need to explicitly progress the communication operations. We show that the communication overhead of parallel application kernels can be reduced up to 92% while not requiring user intervention to make progress. ©2008 IEEE.
@inproceedings{
title = {Optimizing non-blocking collective operations for InfiniBand},
type = {inproceedings},
year = {2008},
keywords = {Application programming interfaces (API); Applicat,Collective operations; Non-blocking,Parallel processing systems},
websites = {https://www.scopus.com/inward/record.uri?eid=2-s2.0-51049098070&doi=10.1109%2FIPDPS.2008.4536138&partnerID=40&md5=774119f7a90a38f066c4bd94fceb5ae7},
city = {Miami, FL},
id = {4f9b684c-49d3-3bc0-8915-42a376a90b26},
created = {2018-01-09T20:30:38.238Z},
file_attached = {false},
profile_id = {42d295c0-0737-38d6-8b43-508cab6ea85d},
last_modified = {2018-03-12T19:03:18.834Z},
read = {false},
starred = {false},
authored = {true},
confirmed = {true},
hidden = {false},
citation_key = {Hoefler2008},
source_type = {conference},
notes = {cited By 14; Conference of IPDPS 2008 - 22nd IEEE International Parallel and Distributed Processing Symposium ; Conference Date: 14 April 2008 Through 18 April 2008; Conference Code:73339},
folder_uuids = {2aba6c14-9027-4f47-8627-0902e1e2342b},
private_publication = {false},
abstract = {Non-blocking collective operations have recently been shown to be a promising complementary approach for overlapping communication and computation in parallel applications. However, in order to maximize the performance and usability of these operations it is important that they progress concurrently with the application without introducing CPU overhead and without requiring explicit user intervention. While studying non-blocking collective operations in the context of our portable library (libNBC), we found that most MPI implementations do not sufficienctly support overlap over the InfiniBand network. To address this issue, we developed a low-level communication layer for libNBC based on the Open Fabrics InfiniBand verbs API. With this layer we are able to achieve high degrees of overlap without the need to explicitly progress the communication operations. We show that the communication overhead of parallel application kernels can be reduced up to 92% while not requiring user intervention to make progress. ©2008 IEEE.},
bibtype = {inproceedings},
author = {Hoefler, T and Lumsdaine, A},
doi = {10.1109/IPDPS.2008.4536138},
booktitle = {IPDPS Miami 2008 - Proceedings of the 22nd IEEE International Parallel and Distributed Processing Symposium, Program and CD-ROM}
}
Downloads: 0
{"_id":"ediGeSdKm229vNNAZ","bibbaseid":"hoefler-lumsdaine-optimizingnonblockingcollectiveoperationsforinfiniband-2008","downloads":0,"creationDate":"2018-03-12T19:10:28.094Z","title":"Optimizing non-blocking collective operations for InfiniBand","author_short":["Hoefler, T.","Lumsdaine, A."],"year":2008,"bibtype":"inproceedings","biburl":"https://bibbase.org/service/mendeley/42d295c0-0737-38d6-8b43-508cab6ea85d","bibdata":{"title":"Optimizing non-blocking collective operations for InfiniBand","type":"inproceedings","year":"2008","keywords":"Application programming interfaces (API); Applicat,Collective operations; Non-blocking,Parallel processing systems","websites":"https://www.scopus.com/inward/record.uri?eid=2-s2.0-51049098070&doi=10.1109%2FIPDPS.2008.4536138&partnerID=40&md5=774119f7a90a38f066c4bd94fceb5ae7","city":"Miami, FL","id":"4f9b684c-49d3-3bc0-8915-42a376a90b26","created":"2018-01-09T20:30:38.238Z","file_attached":false,"profile_id":"42d295c0-0737-38d6-8b43-508cab6ea85d","last_modified":"2018-03-12T19:03:18.834Z","read":false,"starred":false,"authored":"true","confirmed":"true","hidden":false,"citation_key":"Hoefler2008","source_type":"conference","notes":"cited By 14; Conference of IPDPS 2008 - 22nd IEEE International Parallel and Distributed Processing Symposium ; Conference Date: 14 April 2008 Through 18 April 2008; Conference Code:73339","folder_uuids":"2aba6c14-9027-4f47-8627-0902e1e2342b","private_publication":false,"abstract":"Non-blocking collective operations have recently been shown to be a promising complementary approach for overlapping communication and computation in parallel applications. However, in order to maximize the performance and usability of these operations it is important that they progress concurrently with the application without introducing CPU overhead and without requiring explicit user intervention. While studying non-blocking collective operations in the context of our portable library (libNBC), we found that most MPI implementations do not sufficienctly support overlap over the InfiniBand network. To address this issue, we developed a low-level communication layer for libNBC based on the Open Fabrics InfiniBand verbs API. With this layer we are able to achieve high degrees of overlap without the need to explicitly progress the communication operations. We show that the communication overhead of parallel application kernels can be reduced up to 92% while not requiring user intervention to make progress. ©2008 IEEE.","bibtype":"inproceedings","author":"Hoefler, T and Lumsdaine, A","doi":"10.1109/IPDPS.2008.4536138","booktitle":"IPDPS Miami 2008 - Proceedings of the 22nd IEEE International Parallel and Distributed Processing Symposium, Program and CD-ROM","bibtex":"@inproceedings{\n title = {Optimizing non-blocking collective operations for InfiniBand},\n type = {inproceedings},\n year = {2008},\n keywords = {Application programming interfaces (API); Applicat,Collective operations; Non-blocking,Parallel processing systems},\n websites = {https://www.scopus.com/inward/record.uri?eid=2-s2.0-51049098070&doi=10.1109%2FIPDPS.2008.4536138&partnerID=40&md5=774119f7a90a38f066c4bd94fceb5ae7},\n city = {Miami, FL},\n id = {4f9b684c-49d3-3bc0-8915-42a376a90b26},\n created = {2018-01-09T20:30:38.238Z},\n file_attached = {false},\n profile_id = {42d295c0-0737-38d6-8b43-508cab6ea85d},\n last_modified = {2018-03-12T19:03:18.834Z},\n read = {false},\n starred = {false},\n authored = {true},\n confirmed = {true},\n hidden = {false},\n citation_key = {Hoefler2008},\n source_type = {conference},\n notes = {cited By 14; Conference of IPDPS 2008 - 22nd IEEE International Parallel and Distributed Processing Symposium ; Conference Date: 14 April 2008 Through 18 April 2008; Conference Code:73339},\n folder_uuids = {2aba6c14-9027-4f47-8627-0902e1e2342b},\n private_publication = {false},\n abstract = {Non-blocking collective operations have recently been shown to be a promising complementary approach for overlapping communication and computation in parallel applications. However, in order to maximize the performance and usability of these operations it is important that they progress concurrently with the application without introducing CPU overhead and without requiring explicit user intervention. While studying non-blocking collective operations in the context of our portable library (libNBC), we found that most MPI implementations do not sufficienctly support overlap over the InfiniBand network. To address this issue, we developed a low-level communication layer for libNBC based on the Open Fabrics InfiniBand verbs API. With this layer we are able to achieve high degrees of overlap without the need to explicitly progress the communication operations. We show that the communication overhead of parallel application kernels can be reduced up to 92% while not requiring user intervention to make progress. ©2008 IEEE.},\n bibtype = {inproceedings},\n author = {Hoefler, T and Lumsdaine, A},\n doi = {10.1109/IPDPS.2008.4536138},\n booktitle = {IPDPS Miami 2008 - Proceedings of the 22nd IEEE International Parallel and Distributed Processing Symposium, Program and CD-ROM}\n}","author_short":["Hoefler, T.","Lumsdaine, A."],"urls":{"Website":"https://www.scopus.com/inward/record.uri?eid=2-s2.0-51049098070&doi=10.1109%2FIPDPS.2008.4536138&partnerID=40&md5=774119f7a90a38f066c4bd94fceb5ae7"},"biburl":"https://bibbase.org/service/mendeley/42d295c0-0737-38d6-8b43-508cab6ea85d","bibbaseid":"hoefler-lumsdaine-optimizingnonblockingcollectiveoperationsforinfiniband-2008","role":"author","keyword":["Application programming interfaces (API); Applicat","Collective operations; Non-blocking","Parallel processing systems"],"metadata":{"authorlinks":{}},"downloads":0},"search_terms":["optimizing","non","blocking","collective","operations","infiniband","hoefler","lumsdaine"],"keywords":["application programming interfaces (api); applicat","collective operations; non-blocking","parallel processing systems"],"authorIDs":[],"dataSources":["zgahneP4uAjKbudrQ","ya2CyA73rpZseyrZ8","2252seNhipfTmjEBQ"]}