Upgrade to Pro
— share decks privately, control downloads, hide ads and more …
Speaker Deck
Sign up for free
Menu
Search
Features
All features
Private URLs
Password Protection
Custom URLS
Scheduled publishing
Remove Branding
Restrict embedding
Deck Collections
Notes
Features
All features
Private URLs
Password Protection
Custom URLS
Scheduled publishing
Remove Branding
Restrict embedding
Deck Collections
Notes
Explore
Featured decks
Featured speakers
Programming
Technology
Storyboards
Explore
Featured decks
Featured speakers
Programming
Technology
Storyboards
Pricing
Search
Sign in
Sign up for free
ゆううきの研究開発まとめ (2019年2月版) / The summary of yuuki'...
Search
Yuuki Tsubouchi (yuuk1)
February 13, 2019
Research
5.2k
1
Share
Embed
Copy iframe code
Copy JS code
Copy link
Start on current slide
ゆううきの研究開発まとめ (2019年2月版) / The summary of yuuki's research and development in 02/2019
さくらインターネット研究所 研究会 2019.02.13
これまでの研究開発や開発運用について、さくらインターネット研究所内の研究会で話しました。
研究員 坪内佑樹
Yuuki Tsubouchi (yuuk1)
February 13, 2019
More Decks by Yuuki Tsubouchi (yuuk1)
See All by Yuuki Tsubouchi (yuuk1)
「アラーティング」の話をしよう— SREconや論文等の最先端とのギャップをみる
yuukit
7
2.2k
SAKURAONE:An Open Ethernet-based AI HPC System And Its Observed Workload Dynamicsin a Single-Tenant LLM Development Environment
yuukit
1
730
AIスーパーコンピュータにおけるLLM学習処理性能の計測と可観測性 / AI Supercomputer LLM Benchmarking and Observability
yuukit
1
1.3k
SREはサイバネティクスの夢をみるか? / Do SREs Dream of Cybernetics?
yuukit
3
870
SREのためのテレメトリー技術の探究 / Telemetry for SRE
yuukit
14
4.1k
AIスパコン「さくらONE」の オブザーバビリティ / Observability for AI Supercomputer SAKURAONE
yuukit
2
1.6k
AIスパコン「さくらONE」のLLM学習ベンチマークによる性能評価 / SAKURAONE LLM Training Benchmarking
yuukit
2
1.3k
とあるSREの博士「過程」 / A Certain SRE’s Ph.D. Journey
yuukit
11
7.7k
eBPFを用いたAIネットワーク監視システム論文の実装 / eBPF Japan Meetup #4
yuukit
3
1.9k
Other Decks in Research
See All in Research
A new deal for disabled people
resolutionfoundation
0
370
量子サマースクール2026「量子計算機アーキテクチャ分野の概観」
youten622
1
1k
SigLIP2 Explanation
auranightsky
0
140
生成AIなんでも展示会vol6 LT登壇資料 NexteraBERT
rikkabotan7
0
210
Empreinte carbone des ovins : une approche globale (technique, économique, environnementale) – 4 années de résultats avec LIFE GREEN SHEEP
institutdelelevage
PRO
0
150
Physical AIでモデリングはどう変わるか/How Physical AI Will Change Modeling
stktu
0
350
Du suivi individuel à la sélection animale : comment Digi4Live valorise les données des technologies numériques en élevage
institutdelelevage
PRO
0
120
実例から見るLLMのマンガ理解:実務VQAタスクによる長期的文脈と視覚情報の定性評価
kzmssk
0
220
オンライン凸最適化の基礎とゲーム理論への応用 (IBIS 2026 チュートリアル)
taira_tsuchiya
2
1.4k
J-STAGEの現況と全文XML登載必須化について
xspa2012
0
400
[CV勉強会@関東 CVPR2026] PSDesigner: Automated Graphic Design with a Human-Like Creative Workflow / kantocv 67th CVPR 2026
shunk031
0
400
Bilan travail en élevage : première restitution des enquêtes 2025-2026
institutdelelevage
PRO
0
280
Featured
See All Featured
Information Architects: The Missing Link in Design Systems
soysaucechin
1
1.3k
How to Build an AI Search Optimization Roadmap - Criteria and Steps to Take #SEOIRL
aleyda
1
2.2k
The Mindset for Success: Future Career Progression
greggifford
PRO
0
560
The Organizational Zoo: Understanding Human Behavior Agility Through Metaphoric Constructive Conversations (based on the works of Arthur Shelley, Ph.D)
kimpetersen
PRO
0
530
Visualization
eitanlees
153
17k
XXLCSS - How to scale CSS and keep your sanity
sugarenia
250
1.3M
Optimising Largest Contentful Paint
csswizardry
37
4k
Navigating Algorithm Shifts & AI Overviews - #SMXNext
aleyda
1
1.7k
Winning Ecommerce Organic Search in an AI Era - #searchnstuff2025
aleyda
2
2.2k
Put a Button on it: Removing Barriers to Going Fast.
kastner
60
4.6k
Kristin Tynski - Automating Marketing Tasks With AI
techseoconnect
PRO
0
590
Save Time (by Creating Custom Rails Generators)
garrettdimon
PRO
33
5.7k
Transcript
͘͞ΒΠϯλʔωοτ גࣜձࣾ (C) Copyright 1996-2019 SAKURA Internet Inc ͘͞ΒΠϯλʔωοτ ݚڀॴ
Ώ͏͏͖ͷݚڀ։ൃ·ͱΊ (20192݄൛) 2019/02/13 ݚڀһ ௶ ༎थ ͘͞ΒΠϯλʔωοτݚڀॴ ݚڀձ 2019.02.13 @yuuk1t / id:y_uuki
2 1. ݚڀςʔϚ 2. େֶɾେֶӃ࣌ͷݚڀ 3. ΤϯδχΞͱͯ͠ͷݚڀ։ൃ 4. ͘͞ΒͰͷݚڀ։ൃ͍͖ͬͯ 5.
·ͱΊ ΞδΣϯμ
1. ݚڀςʔϚ
4 ਓؒͱίϯϐϡʔλʔͷؔੑͱӡ༻ ίϯϐϡʔλʔ γεςϜ Ϣʔβʔ ։ൃӡ༻ऀ ӡ༻ ӡ༻ • ։ൃӡ༻ऀ͚ͷίϯϐϡʔλʔγεςϜଟ͋Γ
Ϣʔβʔͱ։ൃӡ༻ऀͷ۠ผᐆດʹͳ͍ͬͯΔ • ਓؒͱίϯϐϡʔλʔͷؔੑมԽͯ͠”ӡ༻”͕ ଘࡏ͢Δ͜ͱʹมΘΓͳ͍ “ӡ༻”: มߋ => ҭ => ର ݱ࣮ίϯϐϡʔλʔͷมߋ͕ ා͍
5 ݚڀςʔϚ: ӡ༻Λָ͘͢͠Δ ɾίϯϐϡʔλʔγεςϜͷڪාΛͳ͘͠ӡ༻Λָ͠ΊΔΑ͏ʹ͢ Δ ɾָ͘͠ͳ͍ͱਐ·ͳ͍ ɾίϯϐϡʔλʔ͕ਓͷࣄΛୣ͍ͬͯͬͨ͋ͱʹΔͷͷҰͭʹ ίϯϐϡʔλʔͷӡ༻Λָ͠Ή͜ͱͰͳ͍͔
6 ɾා͞ͷݪҼ৴པੑͷԼ ɾ৴པੑΛ੍ޚ͠ɺมߋΛ࠷େԽ͢Δ ɾมߋͰ͖Δ͜ͱָ͍͠ Site Reliability Engineering (SRE) https://blog.yuuk.io/entry/2019/thinking-sre
7 ɾ؍ଌͱ࣮ݧͷϧʔϓΛߴʹճ͠ɺ৴པੑΛ੍ޚͭͭ͠มߋΛ ߴΊΔγεςϜϏδϣϯ Experimentable Infrastructure
8 ɾมߋ͕͍͠ͷσʔλΛͭγεςϜ ɾ͜Ε͔Βਓ͕ؒੜ͢ΔίϯςϯπΑΓ ػց͕ੜ͢Δσʔλ͕ٸ૿ ɾOSϛυϧΣΞͷϨΠϠͷΈͰ͋ΒΏΔ ཁٻΛຬͨ͢ͷࠔ ɾΞϓϦέʔγϣϯӡ༻ٕज़·Ͱ౿ΈࠐΜ ͩਨ౷߹తͳΞϓϩʔν͕ඞཁ Data-Intensive Applications
2. ֶ෦ɾେֶӃ࣌ͷݚڀ
10 ɾେֶֶ෦࣌ ɾॏෳഉআετϨʔδͷͨΊͷSHA-1ܭࢉγεςϜͷSSE໋ྩʹΑΔߴ εϧʔϓοτԽ ɾେֶӃത࢜લظ՝ఔ࣌ ɾTCP/IPελοΫʹ͓͚ΔνΣοΫαϜܭࢉͷGPUΦϑϩʔσΟϯά 2011~2013
11 ϓϩηοαͷΩϟογϡϝϞϦͱSIMDԋࢉ
12 GPGPUͱLinuxΧʔωϧύέοτॲཧ ɾLinuxΧʔωϧͷTCP/IPελοΫॲཧͷ͏ͪɺTCPϔομͷνΣοΫα ϜܭࢉΛGPUʹ·͔ͤͯεϧʔϓοτΛ্ͤ͞Δ ɾGPUͷϝϞϦ্ʹରσʔλ͕ἧͬͯͳ͍ͱGPUͷ࣮ޮޮ͕͍ ɾύέοτ܊Λ1ͭͷετϦʔϜʹ٧ΊࠐΜͰɺGPU্Ͱ·ͱΊͯܭࢉ ͤ͞Δ ɾ1500όΠτఔͷύέοτͰੑೳ͕ͰͣɺJumbo Frame (7000ό
Πτ͙Β͍) ར༻࣌ʹطଘख๏ΑΓεϧʔϓοτ͕ߴ͘ͳΔ
13 ֶ෦ɾେֶӃ࣌ͷৼΓฦΓ ɾϓϩηοαɺฒྻॲཧɺLinuxΧʔωϧɺTCP/IPʹֶ͍ͭͯͨ ɾ͜ͷ͋ͨΓͷॲཧϞσϧͱ࣮ߦ࣌ؒɺϦιʔεফඅͷഽײ͕֮ͳΜͱ ͳ͘Θ͔͍ͬͯΔͱɺ্ҐͷWebαʔόσʔλϕʔεΛѻ͏্Ͱϒ ϥοΫϘοΫε͕গͳ͘ݟ͑ͨ ɾ࣮ࡍʹଘࡏ͠ͳ͍Λղ͍͍ͯΔΑ͏ͳؾ࣋ͪʹͳΔ͜ͱ͋ͬͨ ɾاۀʹೖͬͯɺ࣮ࣾձͷҙࣝΛ࣮ײ্ͨ͠ͰɺݚڀͰ͖ΔͱΑΓ Αͦ͞͏ͩͱߟ͑࢝Ίͨ
14 ͜ͷ͜Ζͷϒϩά ɾ΄ͦ΅ͦͱॻ͍͍ͯͨϒϩά͕ಡ·ΕΔΑ͏ʹ
3. ΤϯδχΞͱͯ͠ͷݚڀ։ൃ
16 ɾαʔόࢹαʔϏεͷߏஙͱӡ༻ ɾϒϩάαʔϏεͳͲɺ౷తͳWebαʔϏεͷߏஙͱӡ༻ ɾશαʔϏεͷΠϯϑϥج൫(ԾԽϓϩϏδϣχϯάͳͲ)ͷ։ൃ ͱӡ༻ ɾϓϩδΣΫτϚωδϝϯτɺνʔϜϏϧσΟϯάɺϝϯλϦϯά ɾγεςϜϏδϣϯ͔ΒΛղܾ͢ΔϙδςΟϒΞϓϩʔνͷ࣮ફ ܦݧͨ͠։ൃɾӡ༻
17 ࣌ܥྻσʔλϕʔεͷߏஙɾӡ༻ɾ։ൃ https://blog.yuuk.io/entry/high-performance-graphite https://blog.yuuk.io/entry/the-rebuild-of-tsdb-on-cloud
18 ϨΨγʔγεςϜͷӡ༻վળ
19 ݚڀ։ൃ࣮ 1.௶༎थ, TimeFuzeΞʔΩςΫνϟߏ - ॲཧͱσʔλͱλΠϚʔΛҰମԽͨ͠σʔ λύΠϓϥΠϯ, ΣϒγεςϜΞʔΩςΫνϟ(WSA)ݚڀձ ୈ1ճ, 201712݄23
2.௶༎थ, αʔόϨε࣌ʹ͓͚ΔϔςϩδχΞε࣌ܥྻσʔλϕʔεΞʔΩςΫ νϟ, ΣϒγεςϜΞʔΩςΫνϟ(WSA)ݚڀձ ୈ2ճ, 201805݄12 3.௶༎थ,ݹխେ, TCPଓͷʹΑΔ؆ུԽ͞ΕͨωοτϫʔΫґଘؔάϥϑ ͷՄࢹԽج൫, ୈ3ճΣϒγεςϜΞʔΩςΫνϟ(WSA)ݚڀձ, 201811݄17 ɾࠃࠪಡ͖จ ɾޱ಄ൃද ௶༎थ,ࡔேਓ,ᖛా݈,দխ,Ѩ෦ത,দຊ྄հ, HeteroTSDB: ҟछࠞ߹Ωʔό ϦϡʔετΞΛ༻͍ͨࣗಈ֊ԽͷͨΊͷ࣌ܥྻσʔλϕʔεΞʔΩςΫνϟ, Πϯ λʔωοτͱӡ༻ٕज़γϯϙδϜจू, 2018, 7-15 (2018-11-29), Dec 2018
20 ɾࠪಡ͖จ 1݅ ɾֶज़ߨԋ x 3݅ ɾࠃൃද x 34݅ ɾύωϧσΟεΧογϣϯ
x 4݅ ɾڝٕ ISUCON ຊઓग़ 2ճ ɾϒϩάଟ ࣮·ͱΊ
21 ɾ։ൃͨ͠༰Λݚڀͱͯ͠Έ͠ɺจߘݚڀձൃද ɾ࣮ࡍͷࣾձͷཁٻʹج͍ͮͨҙࣝΛѻ͑ɺϓϩμΫγϣϯͰಈ ࡞͍ͯ͠ΔͷΛจʹॻ͚Δ ɾҰํͰɺ࣮Ͱظతͳ༏ઌʹͳΓ͕ͪͳͨΊɺݚڀΛҭͯ ͍ͯ͘͜ͱ͍͠ ɾͰ։ൃͨ͠ͷͰϓϩμΫΫγϣϯڥͰීٴͤ͞Δʹɺ ୯Ґͷ͕͔͔࣌ؒͬͨ ΤϯδχΞͱͯ͠ͷݚڀ։ൃৼΓฦΓ
4. ͘͞ΒͰͷݚڀ։ൃ͍͖ͬͯ
23 ɾͱαʔόʔؒͷԆݮ ɾσʔληϯλʔͷτϥϑΟοΫϘϦϡʔϜͷݮ ϢϏΩλεσʔληϯλʔ ɾ౷తͳͷΣϒΞϓϦέʔγϣϯ ɾϦΞϧλΠϜͳ࣌ܥྻσʔλΛऩू͢ΔΞϓϦέʔγϣϯ Potentials ͕ࣗఆ͢ΔApplications Challenges ɾϢϏΩλεDCڥͰͷσʔλͷஔɺసૹԆɺނোΛಁա͢Δ
24 ɾϢϏΩλεσʔληϯλʔߏͷ͏ͪɺData-IntensiveྖҬΛ୲ ɾԆͷେ͖ͳࢄڥͰࢄγεςϜͱͯ͠ͷಁաੑΛ୲อ͢Δ͜ ͱ͕՝ ɾ͋ΒΏΔΞϓϦέʔγϣϯΛมߋͳ͘ϢϏΩλεDC্ͰޮՌతʹಈ ࡞ͤ͞Δ͜ͱ͍͠ ɾ·ͣɺΞϓϦέʔγϣϯΛԾఆͯ͠ɺঃʑʹҰൠԽ͍ͯ͘͠ ɾWordPressͳͲͷ౷తͳWebΞϓϦέʔγϣϯ ɾIoTͳͲͷϦΞϧλΠϜ࣌ܥྻσʔλऩूΞϓϦέʔγϣϯ ݚڀํ
25 ݚڀϚΠϧετʔϯ: จߘۦಈ จߘ ༰ ~4݄ IOT45 σʔλΠϯςϯγϒɾϢϏΩλεDCͷ ઃܭ ~5݄
IOT46 TimeFuzeΞʔΩςΫνϟ ~7݄ DICOMO2019 (COMPSAC 2019) ࣌ܥྻσʔλϕʔε ~8݄ FIT 2019 άϥϑߏՄࢹԽ ~9݄ IOTS2019(ࠪಡ) ↑ͷதͰࣗ৴ͷ͋Δͷ
26 ɾจͷʮ͡ΊʹʯΛॻ͚Δ͜ͱΛϚΠϧετʔϯͱͯ͠ɺ࠷ॳ͔ ΒετʔϦʔΛͭ͘Γͳ͕Βɺݚڀ։ൃ͍ͯ͘͠ ɾͻͱͭͷςʔϚΛҭͯΔ ɾݚڀձจɺࠃࠪಡ͖จɺࠃࡍձٞɺδϟʔφϧΛ௨ͨ͡ ࣭ͷ্ ɾจͷͨΊͷίʔυͰͳ͘ɺ࣮ફͰ͑ΔιϑτΣΞͱͯ͠࡞ ΓࠐΜͰ͍͘ ࠓޙͷల