Research Paper Introduction #98 "NSDI 2022 recap"

https://www.usenix.org/conference/nsdi22/technical-sessions

Transcript

1. Research Paper Introduction #36 “NSDI 2022 recapͬΆ͍΋ͷ” ௨ࢉ#98 @cafenero_777 2022/05/12
2. Agenda • NSDI 2022঺հ • ؾʹͳΔ࿦จͨͪʢͷ͞ΘΓʣΛ঺հ • 19papers
3. $ which • NSDI 2022 Technical Sessions • Monday, April 4-6, 2022 • https://www.usenix.org/conference/nsdi22/technical-sessions • 78/396 papers, acceptance rate: 19.7% • ॳͷhybrid։࠵ • ॳͷdual-track
4. Awards • Best Paper • Graham: Synchronizing Clocks by Leveraging Local Clock Properties • Community Award • Learning to Communicate E ff ectively Between Battery-free Devices • Packet Order Matters! Improving Application Performance by Deliberately Delaying Packets
5. NSDI ’22 Technical Sessions • 2022/04/04 • Cluster Resource Management • Transport Layer - Part 1 • Video Streaming • Programmable Switches - Part 1 • Security and Privacy • Network Troubleshooting and Debugging • Operational Track - Part 1 • Wireless - Part 1 • 2022/04/06 • Operational Track - Part 2 • Edge IoT Applications • Cloud Scale Services • ISPs and CDNs • Cloud Scale Resource Management • Data Center Network Infrastructure • Multi-tenancy • Software Switching and Beyond • 2022/04/05 • Reliable Distributed Systems • Raising the Bar for Programmable Hardware • Testing and Veri fi cation • Programmable Switches - Part 2 • Sketch-based Telemetry • Transport Layer - Part 2 • Troubleshooting • Wireless - Part 2 24 tracks, 78sessions
6. ࢀߟɿNSDI '19 Technical Sessions • 2019/02/26 • Host Networking • Distributed Systems • Modern Network Hardware • Analytics • Data Center Network Architecture • 2019/02/28 • Network Characterization • Privacy and Security • Network Modeling • Wireless Applications • 2019/02/27 • Wireless Technologies • Operating Systems • Monitoring and Diagnosis • Improving Machine Learning • Network Functions • Wireless Applications 15 tracks, 50sessions
7. ·ͱΊΔํ਑ • ஫ҙ • ʢࢲͷʣڵຯ͕͋ͬͨ΋ͷ͚ͩ঺հ • ʢࢲͷʣཧղͰ͖ͨ΋ͷ͚ͩ঺հ • ۤखͳ΋ͷ: NIC queue, Distributed system, AI/DL, Semantics, Veri fi cation, Compiler, Wireless, Edge/IoT • ͭ·Γɺ͍ͭ΋ͷʢࢲͷʣج४
8. Ξϯέʔτ • ͓΋͠Ζͦ͏ͳ΋ͷΛ3ͭબΜͰΈ͍ͯͩ͘͞ɻ
9. Transport Layer - Part 1
10. PowerTCP: Pushing the Performance Limits of Datacenter Networks University of Vienna • Power (Bandwidth-window product)Λࢦඪͱ͢Δ • QueueΛ୹͘อͪͳ͕Βɺόʔετ/incastʹରԠɻطଘTCPΑΓߴੑೳ
11. FlexTOE: Flexible TCP Of fl oad with Fine-Grained Parallelism University of Washington, UT Austin, MPI-SWS • SmartNIC޲͚TCPΦϑϩʔυΤϯδϯ(TOE) • POSIXιέοτରԠɺx86ൺͰ2.4~4ഒ޲্ • https://github.com/tcp-acceleration-service/FlexTOE
12. Programmable Switches - Part 1
13. SwiSh: Distributed Shared State Abstractions for Programmable Switches Technion, Microsoft Research, The Open University of Israel • P4SWͷͨΊͷ෼ࢄಉظͷ࢓૊ΈΛ࡞Γɺ෼ࢄstatefulσʔλϓϨʔϯΛ࣮૷ • ʢNAT, DDoSݕ஌, rate-limitʣ • Update, replicateੑೳ͕ͱͯ΋ྑ͍
14. Network Troubleshooting and Debugging
15. Closed-loop Network Performance Monitoring and Diagnosis with SpiderMon Rice University, Indian Institute of Technology Hyderabad • ৗ࣌؂ࢹύέοτΛྲྀ͠ɺඞཁͳ͚࣌ͩ෼ੳϨϙʔτΛग़͢ • ௿ΦʔόʔϔουɾߴΧόϨοδ
16. Operational Track - Part 1
17. Decentralized cloud wide-area network traf fi c engineering with BLASTSHIELD Microsoft • B.R.Λ࠷খʹ͠ͳ͕ΒTE͕Մೳͳ
 ରো֐ੑͷߴ͍෼ࢄWANίϯτϩʔϥΛ࡞Δ
18. Bluebird: High-performance SDN for Bare-metal Cloud Services Arista, Intel, Microsoft • AzureͷϕΞϝλϧɾΫϥ΢υαʔϏε༻ͷԾ૝NWΛP4SWͰ·͔ͳ͏ • Netapp, Cray, SAP • 100Gbps, 2೥ӡ༻ • ೔ຊޠղઆهࣄ
19. Cetus: Releasing P4 Programmers from the Chore of Trial and Error Compiling Tsinghua University, Alibaba Group • P4SWͷϦιʔε੍ݶͷͨΊɺιʔεΛίϯύΠϧͰ͖ͳ͍໰୊ • खಈվमͰ͸ͳ͘ɺࣗಈͰม׵͢ΔγεςϜΛ࡞ͬͨ • P4/P4 τϥϯεύΠϥ • ։ൃ͕࣌ؒO(day) -> O(min)
20. Reliable Distributed Systems
21. Graham: Synchronizing Clocks by Leveraging Local Clock Properties Ali Naja fi , Meta; Michael Wei, VMware Research • Awarded Best Paper! • ΫϩοΫಉظΛ͠ͳͯ͘΋ɺϩʔΧϧΫϩοΫ ͷಛੑΛηϯγϯάɾֶशͯ͠ϞσϧԽ͠ɺΫ ϩοΫυϦϑτΛ࠷େ2000ഒվળ • ΫϩοΫಉظ͠ͳ͍ͱ”ͣΕΔ”ɺͱ͍͏ਆ࿩Λ ෷১ɻ௥ՃHW΋ແ͠ͰϚΠΫϩඵਫ਼౓Λҡ࣋
22. Raising the Bar for Programmable Hardware
23. Re-architecting Traf fi c Analysis with Neural Network Interface Cards NEC Laboratories Europe, et-al • B (Binary) NN -> C/P4 -> NICͰτϥϑΟοΫ෼ੳɾҟৗݕ஌ • NFP4000, NetFPGAͰ࣮૷ • wire rate, ௿ϨΠςϯγʔΛ࣮ݱ
24. Elixir: A High-performance and Low-cost Approach to Managing Hardware/ Software Hybrid Flow Tables Considering Flow Burstiness Tsinghua University, Tencent • όʔεττϥϑΟοΫͷHW(P4)/SW(DPDK) ͰΦϑϩʔυʢ͍ΘΏΔ8/2ͷ๏ଇతͳʣ • CPUϦιʔεར༻ͷ࡟ݮͱtail latencyͷվળ
25. Transport Layer - Part 2
26. Packet Order Matters! Improving Application Performance by Deliberately Delaying Packets KTH Royal Institute of Technology Ericsson Research • Community Award Winner! • τϥϑΟοΫͷϩʔΧϦςΟ͕Θ͔ͣͰ΋Լ͕Δͱੑೳେ෯ݮ • ϓϩτίϧɾυϥΠόɾεΠον͕ϩʔΧϦςΟԼ͛Δ • ஗Ԇͤͯ͞Ͱ΋ɺϩʔΧϦςΟΛߴΊΔιϑτ΢ΣΞReframerΛ࡞ͬͨ • WebαʔόͰFCT 11%୹ॖɺεϧʔϓοτ20%޲্ɺ஗Ԇ΋վળ
27. Troubleshooting
28. Buffer-based End-to-end Request Event Monitoring in the Cloud Tsinghua University, Alibaba Group • RLA (Request Latency Anomalies)ΛଌΔBufScopeͷ঺հ • ϦΫΤετIDΛSmartNICͰׂΓৼΓɺ௿஗ԆɺߴΧόϨοδΛ࣮ݱ
29. How to diagnose nanosecond network latencies in rich end-host stacks ETH Zurich, VMware • ௿Φʔόʔϔουͳ஗Ԇ਍அπʔϧ NSightͷ঺հ • طଘπʔϧ20छͱൺֱ • CPU to NICؒͷϝοηʔδΛO(ns)Ͱ࠶ݱ • kernel/user spaceͷ஗ԆݪҼΛൃݟͰ͖ΔɻmemcachedΛ 99.9%ile latencyΛ2.2ms -> 41us • OSSԽ༧ఆ
30. Operational Track - Part 2
31. MLaaS in the Wild: Workload Analysis and Scheduling in Large-Scale Heterogeneous GPU Clusters Hong Kong University of Science and Technology, Alibaba Group • AlibabaͷGPU workload, ར༻཰ͷ௿͞ɺεέδϡʔϦϯάͷݫ͠͞ɺCPUϘτ ϧωοΫͳͲͷ঺հɺ2ϲ݄ͷຊ൪τϨʔεͷղੳ
32. Edge IoT Applications
33. In-Network Velocity Control of Industrial Robot Arms ELTE, Budapest University of Technology and Economics, 3 Ericsson Research • remoteʹ͋ΔP4 switchΛ࢖ͬͯɺϩϘοτΞʔϜΛԁ׈ʹૢ࡞ • يಓ৘ใΛP4lang tableʹຒΊࠐΉ • latency/jitterΛେ෯࡟ݮ
34. Data Center Network Infrastructure
35. Zeta: A Scalable and Robust East-West Communication Framework in Large- Scale Clouds University of Science and Technology of China, 
 Johns Hopkins University, Futurewei Technologies, SUNY at Buffalo • Clos (East-West)௨৴ʹ߹ΘͤͯɺgatewayͰ͸ͳ͘gateway cluster (multi IPs)Λ ༻͍ͨNWઃܭͷ঺հ • ো֐ճ෮͕10ഒɺόʔετϏσΦτϥϑΟοΫͰRTTΛ5.1ഒ@99%ile
36. Aquila: A uni fi ed, low-latency fabric for datacenter networks Google Inc. • ৽͍͠L2ϓϩτίϧɾεΠονɾASICΛ༻͍ͯ40usҎԼɺ1RMAΛ10usҎԼ • 1RMA (RemoteMemoryAccess) protocol, Dragon fl y topology, TiN (ToR-in-NIC)
37. RDC: Energy-Ef fi cient Data Center Network Congestion Relief with Topological Recon fi gurability at the Edge Rice University, Bytedance Inc. • ToR/αʔόؒʹճ࿏ʢᷖճεΠονʣΛೖΕɺϥοΫؒτϥϑΟοΫʹԠͯ͡ϥοΫؒΛ௚݁ͤ͞Δ • 4-10ഒߴ଎Խɺϫοτ͋ͨΓͰ2.4ഒվળ
38. Software Switching and Beyond
39. Tiara: A Scalable and Ef fi cient Hardware Acceleration Architecture for Stateful Layer-4 Load Balancing Hong Kong University of Science and Technology; Chuanxiong Guo, ByteDance • 1Tbps+, 10M fl ow+ͳstateful L4LBΛFPGA + x86Ͱ࣮ݱ • 1.6Tbps, 80M cur-connɺ1.8M CPSͰ4usҎԼͷlatency
40. ײ૝ͨ͠ײ૝ • ͱʹ͔͘ྔଟ͗͢ʂ • Abstract/ConclusionಡΉ͚ͩͰ΋͠ΜͲ͍ • NSDIʹࢀՃͯ͠ɺ·ͱ·ͬͨ࣌ؒͰಡΉʢ·ͱΊΔʣํ͕Α͍ • ࿩୊ʹͳͬͨ΋ͷҎ֎ʹ΋ڵຯΛऒ͔ΕΔ΋ͷ͕͍ͬͺ͍͋ͬͨ • DC, TCP, Operation, P4, ෼ੳܥ, etc
41. EoP