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
k8s_ml_platform.pdf
Search
kuromatsu
February 19, 2019
Technology
3
2.8k
k8s_ml_platform.pdf
Kubernetes Meetup Tokyo #16
https://k8sjp.connpass.com/event/116799/
kuromatsu
February 19, 2019
Tweet
Share
More Decks by kuromatsu
See All by kuromatsu
Kubernetesのカスタマイズポイントのまとめ
kuromt
0
1.8k
Other Decks in Technology
See All in Technology
Entity Framework Core におけるIN句クエリ最適化について
htkym
0
140
松尾研LLM講座2025 応用編Day3「軽量化」 講義資料
aratako
14
4.7k
ペアーズにおけるAIエージェント 基盤とText to SQLツールの紹介
hisamouna
2
1.9k
M&Aで拡大し続けるGENDAのデータ活用を促すためのDatabricks権限管理 / AEON TECH HUB #22
genda
0
300
20251203_AIxIoTビジネス共創ラボ_第4回勉強会_BP山崎.pdf
iotcomjpadmin
0
160
Cloud WAN MCP Serverから考える新しいネットワーク運用 / 20251228 Masaki Okuda
shift_evolve
PRO
0
130
Snowflake Industry Days 2025 Nowcast
takumimukaiyama
0
150
AWSインフルエンサーへの道 / load of AWS Influencer
whisaiyo
0
240
ハッカソンから社内プロダクトへ AIエージェント ko☆shi 開発で学んだ4つの重要要素
leveragestech
0
440
Autonomous Database - Dedicated 技術詳細 / adb-d_technical_detail_jp
oracle4engineer
PRO
5
11k
ソフトウェアエンジニアとAIエンジニアの役割分担についてのある事例
kworkdev
PRO
1
330
Kiro を用いたペアプロのススメ
taikis
4
2.1k
Featured
See All Featured
We Analyzed 250 Million AI Search Results: Here's What I Found
joshbly
0
340
Ethics towards AI in product and experience design
skipperchong
1
150
10 Git Anti Patterns You Should be Aware of
lemiorhan
PRO
659
61k
Why Our Code Smells
bkeepers
PRO
340
58k
Agile that works and the tools we love
rasmusluckow
331
21k
B2B Lead Gen: Tactics, Traps & Triumph
marketingsoph
0
35
Bridging the Design Gap: How Collaborative Modelling removes blockers to flow between stakeholders and teams @FastFlow conf
baasie
0
410
Why Mistakes Are the Best Teachers: Turning Failure into a Pathway for Growth
auna
0
30
XXLCSS - How to scale CSS and keep your sanity
sugarenia
249
1.3M
The Cost Of JavaScript in 2023
addyosmani
55
9.4k
Designing for Performance
lara
610
70k
Evolving SEO for Evolving Search Engines
ryanjones
0
89
Transcript
KubernetesͰػցֶशج൫Λ ߏஙͨ͠ גࣜձࣾ࢜௨ݚڀॴ ਓೳݚڀॴ @kuromt_ 1
ࣗݾհ • TwitterΞΧϯτ: @kuromt_ • ॴଐɿגࣜձࣾ࢜௨ݚڀॴɹਓೳݚڀॴ – ػցֶशΛࢧ͑Δج൫ͷݚڀ – ͚ࣾʹػցֶशج൫Λk8s্ʹߏஙͯ͠2ؒۙ͘ӡ༻த
2
ࠓ͓͢Δ͜ͱ • ֶशίετ͕ߴ͍KubernetesΛӅṭ͠ɺֶशɾਪ ڥΛ࠶ݱՄೳͳػցֶशج൫Λߏஙͨ͠ • ӡ༻ͰಘͨݟQiitaʹެ։ – ʮػցֶशج൫ΛKubernetesͰӡ༻͖ͯͯ͠ʯ 3
KubernetesΛͬͨػցֶशج൫ͷ՝ ϝτϦΫεऩू kubectl ֶ श σ ʔ λ Ϧ
ι ʔε ͷ ֬ ೝ ֶशɺਪ • Kubernetesͷֶशίετ σʔλαΠΤϯςΟετʹ ͱͬͯߴ͗͢Δ • Job, Deployment … • Service, Ingress … • PV, PVC … • ֶशɺਪڥΛ࠶ݱ – Ϟσϧͷਫ਼͕ѱԽͨ͠ ཧ༝ͷݕূ 4
ߏஙͨ͠ج൫ 5
KubernetesΛӅṭ͢ΔಠࣗWebAPIαʔό • ػցֶशͷֶशɾਪͷͨΊʹKubernetesͷػೳΛऔࣺબ͠ ͯAPIΛཧ – ࠷ऴతʹdockerίϚϯυΛୟ͘߹ͱ΄ͱΜͲ͕ࠩͳ͍͍ํͷύϥ ϝʔλʹམͪண͍ͨ • WebAPIαʔόϚχϑΣετΛੜͯ͠kubernetes clientܦ༝
ͰίϯςφΛੜ – KubernetesΛΞοϓάϨʔυͯ͠WebAPIサーバがKubernetesͷAPI ͷมߋΛٵऩ͢ΔͷͰϢʔβύϥϝʔλΛมߋ͢Δඞཁͳ͍ 6
ҙͷֶशɾਪڥΛ࠶ݱ͢ΔΈ • ࠶ݱ͍ͨ͠ڥΛొɾར༻͢ΔAPIΛ༻ҙ – docker-composeΛج൫ͦͷͷʹొ͢Δײ֮ – ڥ໊ͱόʔδϣϯͰҰҙʹܾ·Δ • ొऀҎ֎ొ͞Ε࣮ͨߦڥΛ࠶ར༻Մೳ –
σʔλΛϚϯτ࣮ͨ͠ߦڥͷָ͕ʹͳΓɺモデルの精度の検証や PoCͷελʔτ͕ૉૣ͘ͳͬͨ 7
ߏஙֶͨ͠शɾਪج൫ ϝτϦΫεऩू ֶ श σ ʔ λ Ϧ ι
ʔε ͷ ֬ ೝ ֶशɺਪ • Ϋϥελߏ – ΦϯϓϨϛε – k8s: v1.13.2(HAߏ) – RAM: 560GBʙ1.5TB – GPU: • V100: 20ʙ40ຕ • P100: 8ʙ80ຕ • Ϧιʔε͕ෆ͢ΕNodeΛ Ճͯ͠ରԠ • શମߏCIͰཧ Web API 8
·ͱΊ • ػցֶशج൫Λߏங – KubernetesͰ͋Δ͜ͱΛҙࣝ͠ͳ͍͍ͯ͘ – WebAPIͰֶशɾਪڥΛొɾ࠶ݱͰ͖Δ • PaaSͱͯ͠ج൫͕ҭͪͭͭ͋Δ –
ݱࡏ170ਓҎ্͕ར༻ – KubernetesͩͱΒͣʹ͍ͬͯΔਓଟ͍ • 2,000ίϯςφҎ্Λಉ࣌ʹಈ͔͠ύϥϝʔλ୳ࡧ͢Δ͜ͱ 9