Upgrade to Pro
— share decks privately, control downloads, hide ads and more …
Speaker Deck
Features
Speaker Deck
PRO
Sign in
Sign up for free
Search
Search
Shiva 〜Nextremerをscale upする機械学習環境〜
Search
Chanmoro
December 14, 2017
Technology
0
180
Shiva 〜Nextremerをscale upする機械学習環境〜
Nextremer 社内で運用している機械学習環境 Shiva のご紹介
tag: Docker, Mesos, Marathon, Jupyter, ceph,
Chanmoro
December 14, 2017
Tweet
Share
More Decks by Chanmoro
See All by Chanmoro
LAPRAS クローラーの変遷
chanmoro
3
2.3k
クローラーをテストしよう!
chanmoro
2
1.8k
対話AI on Kubernetes
chanmoro
0
81
グラフ解析で社長の脳内さらす!
chanmoro
0
380
Other Decks in Technology
See All in Technology
Telemetry APIから学ぶGoogle Cloud ObservabilityとOpenTelemetryの現在 / getting-started-telemetry-api-with-google-cloud
k6s4i53rx
0
140
風が吹けばWHOISが使えなくなる~なぜWHOIS・RDAPはサーバー証明書のメール認証に使えなくなったのか~
orangemorishita
15
5.7k
ロールが細分化された組織でSREと協働するインフラエンジニアは何をするか? / SRE Lounge #18
kossykinto
0
210
Amazon GuardDuty での脅威検出:脅威検出の実例から学ぶ
kintotechdev
0
100
Serverless Meetup #21
yoshidashingo
1
120
Oracle Exadata Database Service on Cloud@Customer X11M (ExaDB-C@C) サービス概要
oracle4engineer
PRO
2
6.3k
Kiroでインフラ要件定義~テスト を実施してみた
nagisa53
3
340
UDDのススメ - 拡張版 -
maguroalternative
1
480
Infrastructure as Prompt実装記 〜Bedrock AgentCoreで作る自然言語インフラエージェント〜
yusukeshimizu
1
110
Google Agentspaceを実際に導入した効果と今後の展望
mixi_engineers
PRO
3
410
Rubyの国のPerlMonger
anatofuz
3
730
LLMでAI-OCR、実際どうなの? / llm_ai_ocr_layerx_bet_ai_day_lt
sbrf248
0
460
Featured
See All Featured
YesSQL, Process and Tooling at Scale
rocio
173
14k
Creating an realtime collaboration tool: Agile Flush - .NET Oxford
marcduiker
31
2.2k
Connecting the Dots Between Site Speed, User Experience & Your Business [WebExpo 2025]
tammyeverts
8
440
Fight the Zombie Pattern Library - RWD Summit 2016
marcelosomers
234
17k
Code Reviewing Like a Champion
maltzj
524
40k
Building Better People: How to give real-time feedback that sticks.
wjessup
367
19k
Fireside Chat
paigeccino
38
3.6k
Become a Pro
speakerdeck
PRO
29
5.5k
Gamification - CAS2011
davidbonilla
81
5.4k
Exploring the Power of Turbo Streams & Action Cable | RailsConf2023
kevinliebholz
34
6k
KATA
mclloyd
32
14k
I Don’t Have Time: Getting Over the Fear to Launch Your Podcast
jcasabona
33
2.4k
Transcript
Copyright © Nextremer, inc. All Rights Reserved. 1 Shiva ʙNextremerΛScale
up͢Δػցֶशڥʙ 14th Dec 2017
Copyright © Nextremer, inc. All Rights Reserved. 2 ࣗݾհ ྆֯
ي(ΖͣΈ ͔͖ͣ) גࣜձࣾNextremer ϓϩάϥϚʔ ※ରγεςϜͷݚڀ։ൃΛ͍ͯ͠ΔاۀͰ͢ • ओʹ͍ͬͯΔ͜ͱ • αʔόʔαΠυ։ൃ • ͜͜࠷ۙ৮ٕͬͨज़ • Terraform, Ansible, Packer ← Infrasctucture as Codeʂ • Docker, Kubernetes, Apache Mesos ← ϙʔλϏϦςΟ͍͜͞͏ʂ • Python, Scala, javascript(ES6) ← ؔܕ͏·͔͚͘Δͱؾ͍͍࣋ͪʂ • Finagle, React.js ← ϚΠΫϩαʔϏε͍ͨ͠ʂϦΞΫςΟϒ͍ͨ͠ʂ
Copyright © Nextremer, inc. All Rights Reserved. 3 ࠓ͓͢͠Δ͜ͱ Nextremer
ͷػցֶशڥ Shiva ʹ͍͓ͭͯ͠͠·͢
Copyright © Nextremer, inc. All Rights Reserved. 4 ػցֶशڥ Shiva
1. ΦϑΟεʹ͋ΔGPUϚγϯʹ 2. WebͷμογϡϘʔυ͔Β 3. ઐ༻ͷJupyterΛ্ཱͪ͛ΒΕΔ
Copyright © Nextremer, inc. All Rights Reserved. 5 Shiva͕ղܾ͢Δ՝ ▪
GPUϚγϯΛ͍͍͚ͨͲɾɾɾ • Ϋϥυඅ༻Λؾʹͯ͠ؾܰʹ͑ͳ͍ • ओʹ༧ࢉ͖ͷPJҎ֎ɺτϨʔχϯάͱ͔ɺҋݚڀͱ͔ • GPUϚγϯΛෳϢʔβʔͰڞ༗͢ΔͷΓʹ͍͘ • ϥΠϒϥϦͷڝ߹ɺ୭͕ͬͯΔ͔Θ͔ΒΜ • GPUϚγϯ্Ͱͷڥߏங͕େมͩͬͨΓ • શһ͕ৄ͍͠Θ͚͡Όͳ͍
Copyright © Nextremer, inc. All Rights Reserved. 6 σϞ ※GIFΞχϝʔγϣϯͰ͢
Copyright © Nextremer, inc. All Rights Reserved. 7 Abstract ▪
ϢʔβʔຖͷDockerίϯςφ ▪ શ෦ೖΓڥ & sudoݖݶ ▪ ϗʔϜdirΛશϚγϯͰڞ༗
Copyright © Nextremer, inc. All Rights Reserved. 8 Architecture Overview
Copyright © Nextremer, inc. All Rights Reserved. 9 Architecture Overview
μογϡϘʔυ͔ΒAPIܦ༝Ͱૢ࡞ ɹɾϦιʔεར༻ঢ়گΛऔಘ ɹɾίϯςφͷىಈɾআ
Copyright © Nextremer, inc. All Rights Reserved. 10 Architecture Overview
શͰϗʔϜσΟϨΫτϦΛڞ༗ ɹɾࢄετϨʔδΛߏங ɹɾίϯςφىಈ࣌ʹϚϯτ
Copyright © Nextremer, inc. All Rights Reserved. 11 Architecture Overview
▪ ϢʔβʔͱϙʔτͷϚοϐϯά • http://host/user/kazuki.morozumi • Jupyter͕Ϧεϯ͢ΔϙʔτϥϯμϜ ▪ ίϯςφՔಇঢ়گΛެ։͢ΔAPI ▪ tmpnbΛΧελϚΠζ • https://github.com/jupyter/tmpnb
Copyright © Nextremer, inc. All Rights Reserved. 12 Infrastructure Overview
Copyright © Nextremer, inc. All Rights Reserved. 13 Infrastructure Overview
ίϯςφΦʔέετϨʔγϣϯʹ Mesos + MarathonΛར༻
Copyright © Nextremer, inc. All Rights Reserved. 14 Infrastructure Overview
▪ Mesos+MarathonΛબͨ͠ཧ༝ • ಛఆͷΞϓϦΛσʔϞϯͱͯ͠ಈ͔ͤΕे • αʔϏεσΟεΧόϦͱ͔͍Βͳ͍ • KubernetesΦʔόʔεϖοΫ͗͢Δ • ߏཁૉ͕ଟ͘ෳࡶͰߏஙɺϝϯςφϯε͕େม • ୯७ʹMesos+MarathonΛࢼͯ͠Έ͔ͨͬͨ • ߏங࣌ใ͕͔ͳΓগͳ͔ͬͨ
Copyright © Nextremer, inc. All Rights Reserved. 15 Infrastructure Overview
CPU Intel Core i7-6700K ϝϞϦ 32GB Disk SSD 500GB GPU GTX-1080 ×2ຕ GPUϚγϯεϖοΫ ݟͨେࣄʂ ɹ- ͱΓ͋͑ͣޫΒͤΔ ɹ- ΈΜͳʹݟ͑Δॴʹஔ͘
Copyright © Nextremer, inc. All Rights Reserved. 16 ӡ༻͍ͯͯ͠ىͬͨ͜ ▪
σΟεΫ্ݶ • ֶशσʔλඦGBΛΞοϓϩʔυͨ͠ΒσΟεΫ্ݶʹୡͯ͠ఀࢭ • ίϯςφଆ͔ΒsshfsͰࣗͷPCͷσΟεΫΛϚϯτͯ͠ճආͨ͠ ▪ ίϯςφ͕ͨ࣌͠ͷηʔϑςΟʔωοτ • ։ൃͨ͠ίʔυͷόάͰϗετͷCPUɾϝϞϦΛ৯͍௵͢ • ϏΪφʔͷϢʔβʔ͕͜ͷΑ͏ͳόάΛ࡞Γ͕ͪ • ରԠࡦίϯςφຖʹϦιʔε੍ݶΛ͚ͭΔ • ස͔ͭӨڹͳͷͰಛʹݱঢ়ରԠͤͣ
Copyright © Nextremer, inc. All Rights Reserved. 17 ৄࡉQiitaͷهࣄʂ
Copyright © Nextremer, inc. All Rights Reserved. 18 NextremerͰ͜Μͳ͜ͱΛ͍ͬͯ·͢ WBC؍ઓઐ༻
Line bot
Copyright © Nextremer, inc. All Rights Reserved. 19 NextremerͰ͜Μͳ͜ͱΛ͍ͬͯ·͢
Copyright © Nextremer, inc. All Rights Reserved. 20 NextremerͰ͜Μͳ͜ͱΛ͍ͬͯ·͢ ▪
ฐࣾCEO ҪΛίϐʔͨ͠ରγεςϜ • 2017/10 ୈ8ճରγεςϜγϯϙδϜʹͯσϞൃද
Copyright © Nextremer, inc. All Rights Reserved. 21 NextremerͰ͜Μͳ͜ͱΛ͍ͬͯ·͢