Deep learning emerges as an important new resource-intensive workload and has been successfully applied in computer vision, speech, natural language processing, and so on. Distributed deep learning is becoming a necessity to cope with growing data and model sizes. Its computation is typically characterized by a simple tensor data abstraction to model multi-dimensional matrices, a dataflow graph to model computation, and iterative executions with relatively frequent synchronizations, thereby making it substantially different from Map\/Reduce style distributed big data computation.<\/p>\n
RPC, commonly used as the communication primitive, has been adopted by popular deep learning frameworks such as TensorFlow, which uses gRPC. We show that RPC is suboptimal for distributed deep learning computation, especially on an RDMA-capable network. The tensor abstraction and dataflow graph, coupled with an RDMA network, offers the opportunity to reduce the unnecessary overhead (e.g., memory copy) without sacrificing programmability and generality. In particular, from a data access point of view, a remote machine is abstracted just as a “device” on an RDMA channel, with a simple memory interface for allocating, reading, and writing memory regions. Our graph analyzer looks at both the data flow graph and the tensors to optimize memory allocation and remote data access using this interface. The result is up to 169% improvement against an RPC implementation optimized for RDMA, leading to faster convergence in the training process.<\/p>\n","protected":false},"excerpt":{"rendered":"
Deep learning emerges as an important new resource-intensive workload and has been successfully applied in computer vision, speech, natural language processing, and so on. Distributed deep learning is becoming a necessity to cope with growing data and model sizes. Its computation is typically characterized by a simple tensor data abstraction to model multi-dimensional matrices, a […]<\/p>\n","protected":false},"featured_media":0,"template":"","meta":{"msr-url-field":"","msr-podcast-episode":"","msrModifiedDate":"","msrModifiedDateEnabled":false,"ep_exclude_from_search":false,"_classifai_error":"","footnotes":""},"msr-content-type":[3],"msr-research-highlight":[],"research-area":[13547],"msr-publication-type":[193716],"msr-product-type":[],"msr-focus-area":[],"msr-platform":[],"msr-download-source":[],"msr-locale":[268875],"msr-post-option":[],"msr-field-of-study":[],"msr-conference":[],"msr-journal":[],"msr-impact-theme":[],"msr-pillar":[],"class_list":["post-581590","msr-research-item","type-msr-research-item","status-publish","hentry","msr-research-area-systems-and-networking","msr-locale-en_us"],"msr_publishername":"","msr_edition":"","msr_affiliation":"","msr_published_date":"2019-3-25","msr_host":"","msr_duration":"","msr_version":"","msr_speaker":"","msr_other_contributors":"","msr_booktitle":"","msr_pages_string":"","msr_chapter":"","msr_isbn":"","msr_journal":"","msr_volume":"","msr_number":"","msr_editors":"","msr_series":"","msr_issue":"","msr_organization":"","msr_how_published":"","msr_notes":"","msr_highlight_text":"","msr_release_tracker_id":"","msr_original_fields_of_study":"","msr_download_urls":"","msr_external_url":"","msr_secondary_video_url":"","msr_longbiography":"","msr_microsoftintellectualproperty":1,"msr_main_download":"","msr_publicationurl":"","msr_doi":"","msr_publication_uploader":[{"type":"url","viewUrl":"false","id":"false","title":"https:\/\/dl.acm.org\/citation.cfm?id=3303975","label_id":"243109","label":0}],"msr_related_uploader":[{"type":"file","viewUrl":"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2019\/04\/Fast-Distributed-Deep-Learning-over-RDMA.pptx","id":"594700","title":"fast-distributed-deep-learning-over-rdma-2","label_id":"243115","label":0}],"msr_attachments":[{"id":594700,"url":"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2019\/06\/Fast-Distributed-Deep-Learning-over-RDMA.pptx"}],"msr-author-ordering":[{"type":"user_nicename","value":"Jilong Xue","user_id":36987,"rest_url":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/microsoft-research\/v1\/researchers?person=Jilong Xue"},{"type":"user_nicename","value":"Youshan Miao","user_id":35038,"rest_url":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/microsoft-research\/v1\/researchers?person=Youshan Miao"},{"type":"user_nicename","value":"Cheng Chen","user_id":31458,"rest_url":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/microsoft-research\/v1\/researchers?person=Cheng Chen"},{"type":"user_nicename","value":"Ming Wu","user_id":32960,"rest_url":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/microsoft-research\/v1\/researchers?person=Ming Wu"},{"type":"user_nicename","value":"Lintao Zhang","user_id":32693,"rest_url":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/microsoft-research\/v1\/researchers?person=Lintao Zhang"},{"type":"user_nicename","value":"Lidong Zhou","user_id":32673,"rest_url":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/microsoft-research\/v1\/researchers?person=Lidong Zhou"}],"msr_impact_theme":[],"msr_research_lab":[199560],"msr_event":[],"msr_group":[510017,922377],"msr_project":[555282],"publication":[],"video":[],"download":[],"msr_publication_type":"inproceedings","related_content":{"projects":[{"ID":555282,"post_title":"Deep Learning Compiler and Optimizer","post_name":"deep-learning-compiler-and-optimizer","post_type":"msr-project","post_date":"2018-12-04 18:10:52","post_modified":"2023-07-10 03:41:13","post_status":"publish","permalink":"https:\/\/www.microsoft.com\/en-us\/research\/project\/deep-learning-compiler-and-optimizer\/","post_excerpt":"Project Overview This project aims to build a deep learning compiler and optimizer infrastructure that can provide automatic scalability and efficiency optimization for distributed and local execution.\u00a0 Overall, this stack covers two types of general optimizations: fast distributed training over large-scale servers and efficient local execution on various hardware devices.\u00a0 Currently, our optimizations focus on many different parts of the system stack, such as fast distributed training over RDMA, automatic computation placement across devices, automatic…","_links":{"self":[{"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-project\/555282"}]}}]},"_links":{"self":[{"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-research-item\/581590"}],"collection":[{"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-research-item"}],"about":[{"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/types\/msr-research-item"}],"version-history":[{"count":3,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-research-item\/581590\/revisions"}],"predecessor-version":[{"id":594703,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-research-item\/581590\/revisions\/594703"}],"wp:attachment":[{"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/media?parent=581590"}],"wp:term":[{"taxonomy":"msr-content-type","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-content-type?post=581590"},{"taxonomy":"msr-research-highlight","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-research-highlight?post=581590"},{"taxonomy":"msr-research-area","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/research-area?post=581590"},{"taxonomy":"msr-publication-type","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-publication-type?post=581590"},{"taxonomy":"msr-product-type","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-product-type?post=581590"},{"taxonomy":"msr-focus-area","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-focus-area?post=581590"},{"taxonomy":"msr-platform","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-platform?post=581590"},{"taxonomy":"msr-download-source","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-download-source?post=581590"},{"taxonomy":"msr-locale","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-locale?post=581590"},{"taxonomy":"msr-post-option","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-post-option?post=581590"},{"taxonomy":"msr-field-of-study","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-field-of-study?post=581590"},{"taxonomy":"msr-conference","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-conference?post=581590"},{"taxonomy":"msr-journal","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-journal?post=581590"},{"taxonomy":"msr-impact-theme","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-impact-theme?post=581590"},{"taxonomy":"msr-pillar","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-pillar?post=581590"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}