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
Research Paper Introduction #22 VTrace: Automat...
Search
cafenero_777
May 29, 2021
Technology
220
0
Share
Embed
Copy iframe code
Copy JS code
Copy link
Start on current slide
Research Paper Introduction #22 VTrace: Automatic Diagnostic System for Persistent Packet Loss in Cloud-Scale Overlay Network
https://dl.acm.org/doi/10.1145/3387514.3405851
cafenero_777
May 29, 2021
More Decks by cafenero_777
See All by cafenero_777
#51 “Empowering Azure Storage with RDMA”
cafenero_777
3
580
#49 “Gray Failure: The Achilles’ Heel of Cloud-Scale Systems”
cafenero_777
2
160
#50 “Scalable Hierarchical Aggregation Protocol (SHArP): A Hardware Architecture for Efficient Data Reduction”
cafenero_777
0
170
#33 “Destroying networks for fun (and profit)”
cafenero_777
0
130
#34 “MTPSA: Multi-Tenant Programmable Switches”
cafenero_777
0
100
#37 “Bluebird: High-performance SDN for Bare-metal Cloud Services”
cafenero_777
1
180
#39 “Profiling a warehouse-scale computer”
cafenero_777
0
79
#23 “VFP: A Virtual Switch Platform for Host SDN in the Public Cloud”
cafenero_777
0
300
#24 “Ananta: Cloud Scale Load Balancing”
cafenero_777
0
360
Other Decks in Technology
See All in Technology
プロダクト開発組織の現在地(Ver.2026/07) / product-organization
kaonavi
0
480
ここは地獄!つらい朝会を体験することで、チームとしてのより良い振る舞いに気づくワークショップ / The stand-up meeting from hell in the game industry
scrummasudar
0
310
StepFunctionsとGraphRAGを活用した暗黙知活用のためのRAG基盤
yakumo
0
170
AWS環境のセキュリティ不安を解消した企業事例 ~よくある課題と対策を一挙公開~
asanoharuki
0
170
発表と総括 / Presentations and Summary
ks91
PRO
0
210
そのドキュメント、自動化しませんか?
yuksew
1
440
数値で見る Microsoft MVP 〜Spec Kit と GitHub Copilot Agent で作るデータ可視化ダッシュボード〜
yutakaosada
0
140
Jitera Company Deck
jitera
0
540
AIツールを導入しても生産性はあがらない? カオナビが直面した 3つの壁と乗り越え方。/ Overcoming 3 Barriers to AI-Driven Productivity at kaonavi
kaonavi
0
280
変更し続けられるシステムをどう保つか — AI時代のSSoTという設計原則
kawauso
1
1.2k
それでも、技術なブログを書く理由 #kichijojipm / Why I Still Write Tech Blogs Even Now
shinkufencer
0
850
インシデント事例と パッケージの全量解析に学ぶ ソフトウェアサプライチェーンの守り方 / supply-chain-attack-defense
flatt_security
0
1.1k
Featured
See All Featured
How To Stay Up To Date on Web Technology
chriscoyier
790
250k
svc-hook: hooking system calls on ARM64 by binary rewriting
retrage
2
380
Put a Button on it: Removing Barriers to Going Fast.
kastner
60
4.5k
Building an army of robots
kneath
306
46k
What Being in a Rock Band Can Teach Us About Real World SEO
427marketing
0
1.1k
Cheating the UX When There Is Nothing More to Optimize - PixelPioneers
stephaniewalter
287
14k
WCS-LA-2024
lcolladotor
0
750
Organizational Design Perspectives: An Ontology of Organizational Design Elements
kimpetersen
PRO
1
780
Intergalactic Javascript Robots from Outer Space
tanoku
273
27k
Bridging the Design Gap: How Collaborative Modelling removes blockers to flow between stakeholders and teams @FastFlow conf
baasie
0
620
Discover your Explorer Soul
emna__ayadi
2
1.2k
Test your architecture with Archunit
thirion
1
2.3k
Transcript
Research Paper Introduction #22 “VTrace: Automatic Diagnostic System for Persistent
Packet Loss in Cloud-Scale Overlay Network” ௨ࢉ#72 @cafenero_777 2021/05/27 1
Agenda • ରจ • ֓ཁͱಡ͏ͱͨ͠ཧ༝ 1. INTRODUCTION 2. BACKGROUND 3.
DESIGN AND IMPLEMENTATION 4. EVALUATION 5. DEPLOYMENT EXPERIENCES 6. DISCUSSIONS 7. RELATED WORK 8. CONCLUSION 2
ରจ • VTrace: Automatic Diagnostic System for Persistent Packet Loss
in Cloud- Scale Overlay Network • Chongrong Fang– Haoyu Liu–Mao Miao— Jie Ye— Lei Wang— Wansheng Zhang— Daxiang Kang— Biao Lyv— Peng Cheng– Jiming Chen– • –State Key Laboratory of Industrial Control Technology, Zhejiang University • —Alibaba Group • SIGCOMM 2020 3
֓ཁͱಡ͏ͱͨ͠ཧ༝ • ֓ཁ • ΫϥυεέʔϧͷΦʔόʔϨΠNWͷύέϩεࢹ • VFD (vSwitchͳͲ)ʹࠐΜͰɺసૹঢ়گΛѲ • 20ϲ݄ͷӡ༻ܦݧΛڞ༗
• ಡ͏ͱͨ͠ཧ༝ͱײ • ڊେClosࢹʹʹཱͪͦ͏: pingmesh, Net-bouncerͷ࣍ʁ • ԾNWʹϑΥʔΧε -> ྑ͍ҙຯͰཪΒΕͨ • VPCߏͷࢹͷํ -> ͪΌΜͱͨ͠ઃܭɾγεςϜԽ͕ඞਢ 4
1. INTRODUCTION • ҰൠతͳΫϥυNWͱͦͷ՝ • ཧNWΛநԽͯ͠ςφϯτNWͱͯ͠ʢΦʔόʔϨΠͱͯ͠ʣఏڙ • ཧ/ԾSW (VFD: vSwitch/vRouterߋ৽සൟʂ)ͷ͔ᷮͳΤϥʔ͕࣭Өڹɺtcpdump͡ΌʢཧNWҎ্ʹʣແཧʂ
• VFD͕ലେɾཧࢄɺઃఆɾτϙϩδͷมߋɾSLBʹΑΔϑϩʔಡԽ • ݪҼͷಛఆͷϞσϧԽʢϢʔβͷޡૢ࡞ؚΉʣ • VTrace • ϑϩʔͷ fi rst hop VFDͰύέοτΛϚʔΫɻ֤hopͰmatch/logging͢Δɻ͜ΕΛूΊͯϑϩʔΛ࠶ߏங • ੑೳධՁ͠ɺ2018݄̑ҠߦʹAlibaba Cloudʹల։͠ɺௐࠪ࣌ؒͷେ෯ݮ • ྙཧҰ୴আ֎ ʢѱ༻͢ΔͱτϥϑΟοΫࢹͰ͖ΔͨΊʁʣ 5
2. BACKGROUND • Alibaba Cloud • VXLANͰΦʔόʔϨΠ • ECS (VM),
SLB, VFD (vSW/vRT, ͍͕ͭ͜VXLAN͢Δ) • VPC௨৴ (vSW <-> vSW) • Internetͱͷ௨৴ (vSW <-> vRT) • Ϣʔβͱͷ௨৴ (vSW <-> vRT <-> vBR <-> Ϣʔβ) • VPCؒ௨৴ (vSW <-> vRT … vRT <-> vSW) • ύέϩε • ҰաੑɾܧଓੑɺVFDͷઃఆϛειϑτΣΞόά • Ϣʔβૢ࡞ʹΑΔτϙϩδมߋӨڹ • ଳҬෆɾACLϛεΠϯϑϥোͰͳ͍͕ސ٬Өڹ͋Δɻਖ਼ৗӡ༻͕ͩɻɻ • ૣظʹݕ͍ͨ͠ɿ͖͍͠ͰΞϥʔτൃಈʢଳҬͷٸܹͳԼ99%ile latencyʣ • VM ID, NIC_ID, 5tupleΛͬͯϑϩʔߜΓࠐΈɺVTraceͰஅ 6
3. DESIGN AND IMPLEMENTATION (Overview) • શͳpacket traceͱݪҼVFDՕॴಛఆ͕ඞਢཁ݅ • ΦʔόʔϨΠNW্Ͱͷϑϩʔநग़
• ςφϯτ௨৴Өڹͳ͠ɺ͔ͭɺςφϯτଆ࣮ͳ͠ • ΫϥυεέʔϧͳઃܭʢଟͷϑϩʔɾλεΫɺԠੑʣ • VTrace֓ཁɿ2ஈ֊ͰΔ • VTrace Data Generation (VDG) • ֤VFDઃఆɾλʔήοτͷύέοτॲཧ • ϩάग़ྗɾΤʔδΣϯτసૹ • Data Collection and Analysis (DCA) • JStorm (Streamingॲཧ)ͰԾϑϩʔύεͷߏஙɾੳ • taskཧɺ݁ՌཧɺAPI࿈ܞ 7
3. DESIGN AND IMPLEMENTATION VDG: VTrace Data Generation • ଞͷํ๏ɾྨࣅྫ
• SDNతʹίϯτϩʔϥϝοηʔδૹΔɺΦʔόʔϔουɺin/out • RSPANతΞϓϩʔνɺϛϥʔ(dump)͚ͩݟͯݪҼ͕͔Βͳ͍ • ݪҼϊʔυಛఆ͕ओ؟ɺݪҼಛఆ·ͰࣗಈʢʁʣͰͬͯ͘Εͳ͍ • ઃܭ • fast path-slow path • ·ͣslow pathॲཧɺͦͷޙmatch/action rule update (fast path)ɺcoloring/matching/logging • coloring: ͲͷVM௨৴ΛԿύέοτऔಘ͢Δ͔ΛvFWʹઃఆɻϨʔτ্ݶ͖ɻ400 probes. • ࣮ 1. coloring: DSCPΛϚʔΫɺ࣍ͷVFWͰ͏ɻϢχʔΫͳPacket_ID༩ɻରͷશvFDʹdeploy 2. match/action͢ΔɻରFDͷΈઃఆ͞ΕΔɻ~ेϧʔϧఔɻ 3. ֤MetricsऔಘɺNATରԠɺdscp_mark, ϩάసૹ 8
3. DESIGN AND IMPLEMENTATION DCA: Data Collection and Analysis •
ઃܭɾ࣮ • JStorm (Apache Storm based) ετϦʔϜϓϩηεΤϯδϯ • task/log spouts -> ֤bolts (ฒྻॲཧ) -> ݁ՌΛDB • LogूܭԆ͕͋ΔΛߟྀ (trace_time, trace_time, ྫ֎ॲཧ) • Filter/Transfer/Group • 5tuple+NAT_key fi lter, Packet_IDͷ౷ҰԽʢendiannessʣ, TSͰͳ͘ packet_ID, node_ipͰgrouping • Path࠶ߏங • dscp_mark=1, loc:in͕࠷ॳɺ out_dip,out_sip͕มΘΒͳ͚Εಉ͡node, ࠷ޙdscp_mark=2 • Log SpoutsࢄͰෛՙࢄ 9 σʔλੜ σʔλॲཧ ʢલॲཧʣ σʔλॲཧ
4. EVALUATION • ςετڥʢ35+2vRouterʣͱ࣮ࡍͷΫϥυڥʢ90%ͷvFDVTraceಋೖࡁΈʣͰධՁ • Өڹڐ༰ൣғ • vRT: rateͰ14%ఔɺCPUͰ7%ఔʣ//ӈਤ •
vSW: 4%~18%(શύέοτରͷ߹) • σʔλΦʔόʔϔου • Ұ࣌อ࣋ྖҬɿ7.2GB (= 600B*2 * 4hop * 300packet * 5000task)ఔ • ଳҬతʹແࢹͰ͖Δ • ࣮ߦ • ςετڥɿ800task (4hop, 100packet) • 45ඵҎʹtaskྃɺ࠶ߏஙॲཧ0.54msఔ • ຊ൪Ϋϥυڥ • 1packetॲཧʹ5msఔɻλεΫྃʹʙे͔͔Δʢͭ·Γσʔλੜ͕࣌ؒࢧతʣ • खಈͰͷΓ͚࡞ۀʢ࣌ؒʣʹൺΔͱര 10
5. DEPLOYMENT EXPERIENCES • 2018/05ʹAlibaba Cloud্ͷ90%ͷVFDʹVTraceΛల։ޙͷܦݧ • 2018/05: ϢʔβىҼͷΦʔόʔϨΠNWো •
SLBΛܦ༝ͨ͠Πϯλʔωοτ௨৴͕Ͱ͖ͳ͍ʂ • VTraceͰvSwitchloc=in,loc=errorΛൃݟɻݪҼϢʔβACLઃఆʢͭ·ΓΠϯϑϥࣗମਖ਼ৗʣ • 2018/09: ཧNWোͷݕ • ϥετϗοϓ͕ϥετϗοϓ͡Όͳ͍ʂ(dscp_mark!=2, loc=out) • ݪҼNICϦιʔεෆʢετϨʔδ௨৴ʹ৯ΘΕ͍ͯͨʣ • ݟ • VFDؒͷʢΦʔόʔϨΠͷʣlatencyҟৗݕ͍͢͠ • ਖ਼ৗੑͷ͕֬ߴ͍ͨΊΤεύʔ͍͢͠ɿྫɿ ଳҬઃఆʹ͕͋Δ͔ʁ • औಘύέοτɺ࣌ؒΘ͔ͣͰे • ϏδωεϢʔεέʔεͱ࿈ಈʢύέϩεͱௐࠪґཔͷ૿Ճʣ • ࠜຊݪҼͷຆͲϢʔβʹΑΔςφϯτૢ࡞ɻɻɻ 11
6. DISCUSSIONS • ཧNWো݁ߏ͔Δʢલทʣ • VFDؒ௨৴͕͓͔͍͠ -> PFDͷpathΛ࠶ߏͰ͖Δ͔ʁ • Ұ࣌తͳύέϩεݪҼΛΓ͍ͨ
• ࣌ؒΩϟϓνϟݫ͍͠ɻͦͷΘΓʹ౷ܭʢVTrace-statsʣΛ͑Δ͔ʁ • statsͦΕͳΓʹෛՙ͕͔͔ΔΒ͍͠ 12 Physical: ཧ
7. RELATED WORK • Active NW monitoring • pingmesh, ATPG,
Cisco IPSLA: ຊ൪ͱҧ͏ܦ࿏Λ௨ΔՄೳੑ͕͋Γ • Passive NW monitoring • VeriFlowɿΤϯυϗετʹಋೖ͕ඞཁ • Network debugging method • ৭ʑ͋Δ͕ɺݪҼϊʔυಛఆʹࢸΒͳ͍ʢͱ͍͏ओுʣ • INT (In-band Network Telemetry)ലେͳτϥϑΟοΫ/many hopڥͩͱ࣮༻Ṗ 13
CONCLUSION • VTrace • ΫϥυεέʔϧͳύέϩεࣗಈݕγεςϜ • VFDͰcoloring/matching/logging͠ɺରϑϩʔΛௐࠪ • 20ϲ݄Ҏ্Քಇ͠ɺՌग़͍ͯΔ •
ͰஅͰ͖ΔɻԾɾཧΘͣτϥϒϧγϡʔςΟϯάͰ͖ͨ 14