{"id":1140375,"date":"2025-05-27T13:59:39","date_gmt":"2025-05-27T20:59:39","guid":{"rendered":"https:\/\/www.microsoft.com\/en-us\/research\/?post_type=msr-research-item&p=1140375"},"modified":"2025-05-27T13:59:40","modified_gmt":"2025-05-27T20:59:40","slug":"netherite-efficient-execution-of-serverless-workflows-extended-journal-version","status":"publish","type":"msr-research-item","link":"https:\/\/www.microsoft.com\/en-us\/research\/publication\/netherite-efficient-execution-of-serverless-workflows-extended-journal-version\/","title":{"rendered":"Netherite: Efficient Execution of Serverless Workflows (Extended Journal Version)"},"content":{"rendered":"
Serv<\/span>erless<\/span> applications<\/span> are<\/span> popular<\/span> because<\/span> the<\/span>y<\/span> can<\/span> provide<\/span> scalability<\/span> and<\/span> load-based<\/span> billing<\/span> with<\/span> minimal<\/span> de<\/span>v<\/span>eloper<\/span> ef<\/span>fort.<\/span><\/div>\n
Be<\/span>yond<\/span> simple<\/span> stateless<\/span> functions<\/span> (F<\/span>aaS),<\/span> modern<\/span> serv<\/span>erless<\/span> programming<\/span> models<\/span> such<\/span> as<\/span> Durable<\/span> Functions<\/span> (DF)<\/span> no<\/span>w<\/span><\/div>\n
pro<\/span>vide<\/span> stateful<\/span> abstractions,<\/span> such<\/span> as<\/span> workflo<\/span>ws,<\/span> tasks<\/span> and<\/span> actors,<\/span> whose<\/span> state<\/span> and<\/span> progress<\/span> is<\/span> implicitly<\/span> and<\/span> continuously<\/span><\/div>\n
persisted.<\/span> This<\/span> automatic<\/span> resilience<\/span> greatly<\/span> simplifies<\/span> de<\/span>velopment,<\/span> b<\/span>ut<\/span> also<\/span> creates<\/span> performance<\/span> challenges,<\/span> such<\/span> as<\/span> a<\/span> lar<\/span>ge<\/span><\/div>\n
number<\/span> of<\/span> IOps.<\/span> T<\/span>o<\/span> address<\/span> these<\/span> challenges,<\/span> we<\/span> introduce<\/span> Netherite,<\/span> a<\/span> no<\/span>v<\/span>el<\/span> architecture<\/span> for<\/span> ex<\/span>ecuting<\/span> serve<\/span>rless<\/span> w<\/span>orkflo<\/span>ws<\/span><\/div>\n
on<\/span> an<\/span> elastic<\/span> cluster<\/span>.<\/span> Netherite<\/span> groups<\/span> the<\/span> numerous<\/span> application<\/span> objects<\/span> into<\/span> a<\/span> smaller<\/span> number<\/span> of<\/span> partitions,<\/span> and<\/span> pipelines<\/span><\/div>\n
the<\/span> state<\/span> persistence<\/span> of<\/span> each<\/span> partition.<\/span> This<\/span> improv<\/span>es<\/span> latenc<\/span>y<\/span> and<\/span> throughput,<\/span> as<\/span> it<\/span> enables<\/span> workflo<\/span>w<\/span> steps<\/span> to<\/span> group<\/span> commit,<\/span><\/div>\n
e<\/span>ven<\/span> if<\/span> causally<\/span> dependent.<\/span> Moreo<\/span>v<\/span>er<\/span>,<\/span> Netherite<\/span> le<\/span>verages<\/span> F<\/span>ASTER\u2019<\/span>s<\/span> h<\/span>ybrid<\/span> log<\/span> approach<\/span> to<\/span> support<\/span> lar<\/span>ger-than-memory<\/span><\/div>\n
application<\/span> state,<\/span> and<\/span> to<\/span> enable<\/span> ef<\/span>ficient<\/span> partition<\/span> mov<\/span>ement<\/span> between<\/span> compute<\/span> hosts.<\/span> Our<\/span> e<\/span>v<\/span>aluation<\/span> sho<\/span>ws<\/span> that<\/span> (a)<\/span> Netherite<\/span><\/div>\n
achie<\/span>ves<\/span> lo<\/span>wer<\/span> latenc<\/span>y<\/span> and<\/span> higher<\/span> throughput<\/span> than<\/span> the<\/span> original<\/span> DF<\/span> engine,<\/span> by<\/span> more<\/span> than<\/span> an<\/span> order<\/span> of<\/span> magnitude<\/span> in<\/span> some<\/span> cases,<\/span><\/div>\n
and<\/span> (b)<\/span> that<\/span> Netherite<\/span> has<\/span> lo<\/span>wer<\/span> latency<\/span> than<\/span> some<\/span> commonly<\/span> used<\/span> alternati<\/span>v<\/span>es,<\/span> like<\/span> A<\/span>WS<\/span> Step<\/span> Functions<\/span> or<\/span> cloud<\/span> storage<\/span><\/div>\n
triggers.<\/span><\/div>\n","protected":false},"excerpt":{"rendered":"

Serverless applications are popular because they can provide scalability and load-based billing with minimal developer effort. Beyond simple stateless functions (FaaS), modern serverless programming models such as Durable Functions (DF) now provide stateful abstractions, such as workflows, tasks and actors, whose state and progress is implicitly and continuously persisted. This automatic resilience greatly simplifies development, […]<\/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":"","msr-author-ordering":null,"msr_publishername":"","msr_publisher_other":"","msr_booktitle":"","msr_chapter":"","msr_edition":"","msr_editors":"","msr_how_published":"","msr_isbn":"","msr_issue":"","msr_journal":"","msr_number":"","msr_organization":"","msr_pages_string":"","msr_page_range_start":"25","msr_page_range_end":"","msr_series":"","msr_volume":"34","msr_copyright":"","msr_conference_name":"","msr_doi":"","msr_arxiv_id":"","msr_s2_paper_id":"","msr_mag_id":"","msr_pubmed_id":"","msr_other_authors":"","msr_other_contributors":"","msr_speaker":"","msr_award":"","msr_affiliation":"","msr_institution":"","msr_host":"","msr_version":"","msr_duration":"","msr_original_fields_of_study":null,"msr_release_tracker_id":"","msr_s2_match_type":"","msr_citation_count_updated":"","msr_published_date":"2025-2-20","msr_highlight_text":"","msr_notes":"Special Issue \"Best of VLDB 2022\"","msr_longbiography":"","msr_publicationurl":"","msr_external_url":"","msr_secondary_video_url":"","msr_conference_url":"","msr_journal_url":"","msr_s2_pdf_url":"","msr_year":0,"msr_citation_count":0,"msr_influential_citations":0,"msr_reference_count":0,"msr_s2_match_confidence":0,"msr_microsoftintellectualproperty":true,"msr_s2_open_access":false,"msr_s2_author_ids":[],"msr_pub_ids":[],"msr_hide_image_in_river":null,"footnotes":""},"msr-research-highlight":[],"research-area":[13563,13547],"msr-publication-type":[193715],"msr-publisher":[],"msr-focus-area":[],"msr-locale":[268875],"msr-post-option":[269148,269142],"msr-field-of-study":[246691],"msr-conference":[],"msr-journal":[269950],"msr-impact-theme":[],"msr-pillar":[],"class_list":["post-1140375","msr-research-item","type-msr-research-item","status-publish","hentry","msr-research-area-data-platform-analytics","msr-research-area-systems-and-networking","msr-locale-en_us","msr-post-option-approved-for-river","msr-post-option-include-in-river","msr-field-of-study-computer-science"],"msr_publishername":"","msr_edition":"","msr_affiliation":"","msr_published_date":"2025-2-20","msr_host":"","msr_duration":"","msr_version":"","msr_speaker":"","msr_other_contributors":"","msr_booktitle":"","msr_pages_string":"","msr_chapter":"","msr_isbn":"","msr_journal":"","msr_volume":"34","msr_number":"","msr_editors":"","msr_series":"","msr_issue":"","msr_organization":"","msr_how_published":"","msr_notes":"Special Issue \"Best of VLDB 2022\"","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":"doi","viewUrl":"false","id":"false","title":"https:\/\/doi.org\/10.1007\/s00778-024-00898-1","label_id":"243106","label":0},{"type":"url","viewUrl":"false","id":"false","title":"https:\/\/rdcu.be\/eaPT6","label_id":"243109","label":0}],"msr_related_uploader":"","msr_citation_count":0,"msr_citation_count_updated":"","msr_s2_paper_id":"","msr_influential_citations":0,"msr_reference_count":0,"msr_arxiv_id":"","msr_s2_author_ids":[],"msr_s2_open_access":false,"msr_s2_pdf_url":null,"msr_attachments":[],"msr-author-ordering":[{"type":"user_nicename","value":"Sebastian Burckhardt","user_id":33544,"rest_url":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/microsoft-research\/v1\/researchers?person=Sebastian Burckhardt"},{"type":"user_nicename","value":"Badrish Chandramouli","user_id":31166,"rest_url":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/microsoft-research\/v1\/researchers?person=Badrish Chandramouli"},{"type":"text","value":"Chris Gillum (cgillum)","user_id":0,"rest_url":false},{"type":"text","value":"David Justo (dajusto)","user_id":0,"rest_url":false},{"type":"text","value":"Konstantinos Kallas","user_id":0,"rest_url":false},{"type":"text","value":"Connor McMahon","user_id":0,"rest_url":false},{"type":"text","value":"Christopher S. Meiklejohn","user_id":0,"rest_url":false},{"type":"text","value":"Xiangfeng Zhu","user_id":0,"rest_url":false}],"msr_impact_theme":[],"msr_research_lab":[199565],"msr_event":[],"msr_group":[144812,957177],"msr_project":[],"publication":[],"video":[],"msr-tool":[],"msr_publication_type":"article","related_content":[],"_links":{"self":[{"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-research-item\/1140375","targetHints":{"allow":["GET"]}}],"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\/1140375\/revisions"}],"predecessor-version":[{"id":1140386,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-research-item\/1140375\/revisions\/1140386"}],"wp:attachment":[{"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/media?parent=1140375"}],"wp:term":[{"taxonomy":"msr-research-highlight","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-research-highlight?post=1140375"},{"taxonomy":"msr-research-area","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/research-area?post=1140375"},{"taxonomy":"msr-publication-type","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-publication-type?post=1140375"},{"taxonomy":"msr-publisher","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-publisher?post=1140375"},{"taxonomy":"msr-focus-area","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-focus-area?post=1140375"},{"taxonomy":"msr-locale","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-locale?post=1140375"},{"taxonomy":"msr-post-option","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-post-option?post=1140375"},{"taxonomy":"msr-field-of-study","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-field-of-study?post=1140375"},{"taxonomy":"msr-conference","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-conference?post=1140375"},{"taxonomy":"msr-journal","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-journal?post=1140375"},{"taxonomy":"msr-impact-theme","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-impact-theme?post=1140375"},{"taxonomy":"msr-pillar","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-pillar?post=1140375"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}