{"id":581977,"date":"2019-04-26T20:34:33","date_gmt":"2019-04-27T03:34:33","guid":{"rendered":"https:\/\/www.microsoft.com\/en-us\/research\/?post_type=msr-research-item&p=581977"},"modified":"2020-06-11T14:44:23","modified_gmt":"2020-06-11T21:44:23","slug":"mllib-fast-training-of-glms-using-spark-mllib","status":"publish","type":"msr-research-item","link":"https:\/\/www.microsoft.com\/en-us\/research\/publication\/mllib-fast-training-of-glms-using-spark-mllib\/","title":{"rendered":"MLlib*: Fast Training of GLMs using Spark MLlib"},"content":{"rendered":"
In Tencent Inc., more than 80% of the data are
\nextracted and transformed using Spark. However, the commonly
\nused machine learning systems are TensorFlow, XGBoost, and
\nAngel, whereas Spark MLlib, an official Spark package for
\nmachine learning, is seldom used. One reason for this ignorance
\nis that it is generally believed that Spark is slow when it comes
\nto distributed machine learning. Users therefore have to undergo
\nthe painful procedure of moving data in and out of Spark. The
\nquestion why Spark is slow, however, remains elusive.<\/p>\n
In this paper, we study the performance of MLlib with a focus
\non training generalized linear models using gradient descent.
\nBased on a detailed examination, we identify two bottlenecks in
\nMLlib, i.e., pattern of model update and pattern of communication.
\nTo address these two bottlenecks, we tweak the implementation
\nof MLlib with two state-of-the-art and well-known techniques,
\nmodel averaging and AllReduce. We show that, the new system
\nthat we call MLlib*, can significantly improve over MLlib and
\nachieve similar or even better performance than other specialized
\ndistributed machine learning systems (such as Petuum and
\nAngel), on both public and Tencent\u2019s workloads.<\/p>\n","protected":false},"excerpt":{"rendered":"
In Tencent Inc., more than 80% of the data are extracted and transformed using Spark. However, the commonly used machine learning systems are TensorFlow, XGBoost, and Angel, whereas Spark MLlib, an official Spark package for machine learning, is seldom used. One reason for this ignorance is that it is generally believed that Spark is slow […]<\/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":[13563],"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-581977","msr-research-item","type-msr-research-item","status-publish","hentry","msr-research-area-data-platform-analytics","msr-locale-en_us"],"msr_publishername":"","msr_edition":"","msr_affiliation":"","msr_published_date":"2019-4-1","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:\/\/conferences.computer.org\/icde\/2019\/pdfs\/ICDE2019-21DqmQqM7YlfcW2ZceNbB3\/4KpGB88qJHNxzNpagG8ku1\/53MniahSV4B7G0KTsAfB1P.pdf","label_id":"243109","label":0}],"msr_related_uploader":"","msr_attachments":[],"msr-author-ordering":[{"type":"text","value":"Zhipeng Zhang","user_id":0,"rest_url":false},{"type":"text","value":"Jiawei Jiang","user_id":0,"rest_url":false},{"type":"user_nicename","value":"Wentao Wu","user_id":34824,"rest_url":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/microsoft-research\/v1\/researchers?person=Wentao Wu"},{"type":"text","value":"Ce Zhang","user_id":0,"rest_url":false},{"type":"text","value":"Lele Yu","user_id":0,"rest_url":false},{"type":"text","value":"Bin Cui","user_id":0,"rest_url":false}],"msr_impact_theme":[],"msr_research_lab":[199565],"msr_event":[],"msr_group":[957177],"msr_project":[],"publication":[],"video":[],"download":[],"msr_publication_type":"inproceedings","related_content":[],"_links":{"self":[{"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-research-item\/581977"}],"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":1,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-research-item\/581977\/revisions"}],"predecessor-version":[{"id":581980,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-research-item\/581977\/revisions\/581980"}],"wp:attachment":[{"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/media?parent=581977"}],"wp:term":[{"taxonomy":"msr-content-type","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-content-type?post=581977"},{"taxonomy":"msr-research-highlight","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-research-highlight?post=581977"},{"taxonomy":"msr-research-area","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/research-area?post=581977"},{"taxonomy":"msr-publication-type","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-publication-type?post=581977"},{"taxonomy":"msr-product-type","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-product-type?post=581977"},{"taxonomy":"msr-focus-area","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-focus-area?post=581977"},{"taxonomy":"msr-platform","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-platform?post=581977"},{"taxonomy":"msr-download-source","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-download-source?post=581977"},{"taxonomy":"msr-locale","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-locale?post=581977"},{"taxonomy":"msr-post-option","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-post-option?post=581977"},{"taxonomy":"msr-field-of-study","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-field-of-study?post=581977"},{"taxonomy":"msr-conference","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-conference?post=581977"},{"taxonomy":"msr-journal","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-journal?post=581977"},{"taxonomy":"msr-impact-theme","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-impact-theme?post=581977"},{"taxonomy":"msr-pillar","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-pillar?post=581977"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}