devoflow scaling flow management for high performance
play

DevoFlow: Scaling Flow Management for High-Performance Networks - PowerPoint PPT Presentation

Oct 01, 2023 •964 likes •1.97k views

DevoFlow: Scaling Flow Management for High-Performance Networks Andy Curtis Je ff Mogul Jean Tourrilhes Praveen Yalagandula Puneet Sharma Sujata Banerjee Wednesday, August 17, 11 Software-de fi ned networking Wednesday, August 17, 11

  1. DevoFlow: Scaling Flow Management for High-Performance Networks Andy Curtis Je ff Mogul Jean Tourrilhes Praveen Yalagandula Puneet Sharma Sujata Banerjee Wednesday, August 17, 11

  2. Software-de fi ned networking Wednesday, August 17, 11

  3. Software-de fi ned networking • Enables programmable networks Wednesday, August 17, 11

  4. Software-de fi ned networking • Enables programmable networks • Implemented by OpenFlow Wednesday, August 17, 11

  5. Software-de fi ned networking • Enables programmable networks • Implemented by OpenFlow • OpenFlow is a great concept, but... - its original design imposes excessive overheads Wednesday, August 17, 11

  6. Traditional switch Control-plane Data-plane Wednesday, August 17, 11

  7. Traditional switch Control-plane Routed packets Inbound Data-plane packets Wednesday, August 17, 11

  8. Traditional switch Reachability Reachability Control-plane Routed packets Inbound Data-plane packets Wednesday, August 17, 11

  9. Control-plane Centralized controller OpenFlow switch Inbound packets Routed packets Data-plane Wednesday, August 17, 11

  10. Control-plane Centralized controller Flow setups Forwarding table entries Link state Statistics requests Forwarding rule stats Inbound packets Routed packets Data-plane Wednesday, August 17, 11

  11. OpenFlow enables innovative management solutions Wednesday, August 17, 11

  12. OpenFlow enables innovative management solutions • Consistent routing and security policy enforcement [Ethane, SIGCOMM 2007] • Data center network architectures like VL2 and PortLand [Tavakoli et al. Hotnets 2009] • Client load-balancing with commodity switches [Aster*x, ACLD demo 2010; Wang et al., HotICE 2011] • Flow scheduling [Hedera, NSDI 2010] • Energy-proportional networking [ElasticTree, NSDI 2010] • Automated data center QoS [Kim et al., INM/WREN 2010] Wednesday, August 17, 11

  13. But OpenFlow is not perfect... • Scaling these solutions to data center- sized networks is challenging Wednesday, August 17, 11

  14. Contributions • Characterize overheads of implementing OpenFlow in hardware • Propose DevoFlow to enable cost- e ff ective, scalable fl ow management • Evaluate DevoFlow by applying it to data center fl ow scheduling Wednesday, August 17, 11

  15. Contributions • Characterize overheads of implementing OpenFlow in hardware Experience drawn from implementing OpenFlow on HP ProCurve switches Wednesday, August 17, 11

  16. OpenFlow couples fl ow setup with visibility Central controller . . . Core Aggregation . . . Edge switches Wednesday, August 17, 11

  17. Central controller . . . Core Aggregation . . . Edge switches Flow arrival Wednesday, August 17, 11

  18. If no forwarding table rule at switch (exact-match or wildcard) Central controller . . . Core Aggregation . . . Edge switches Wednesday, August 17, 11

  19. Two problems arise... Central controller . . . Core Aggregation . . . Edge switches Wednesday, August 17, 11

  20. problem 1: bottleneck at controller Central controller . . . . . . Wednesday, August 17, 11

  21. problem 1: bottleneck at controller Up to 10 million new fl ows per second in data center with 100 edge switches [Benson et al. IMC 2010] Central controller . . . . . . Wednesday, August 17, 11

  22. problem 1: bottleneck at controller Up to 10 million new fl ows per second in data center with 100 edge switches [Benson et al. IMC 2010] If controller can handle 30K fl ow setups/ sec. then, we need at least 333 controllers! Central controller . . . . . . Wednesday, August 17, 11

  23. Onix [Koponen et al. OSDI 2010] Maestro [Cai et al. Tech Report 2010] HyperFlow [Tootoonchian and Ganjali, WREN 2010] Devolved controller [Tam et al. WCC 2011] Central controller . . . . . . Wednesday, August 17, 11

  24. problem 2: stress on switch control-plane Central controller . . . . . . Wednesday, August 17, 11

  25. Control-plane Inbound packets Routed packets Data-plane Wednesday, August 17, 11

  26. Control-plane Switch Switch CPU control-plane Inbound packets Routed packets ASIC Data-plane Wednesday, August 17, 11

  27. Control-plane Switch Switch CPU control-plane Inbound packets Routed packets ASIC Data-plane Wednesday, August 17, 11

  28. Scaling problem: switches • Inherent overheads • Implementation-imposed overheads Wednesday, August 17, 11

  29. Scaling problem: switches • Inherent overheads - Bandwidth • OpenFlow creates too much control tra ffi c ~1 control packet for every 2–3 data packets - Latency • Implementation-imposed overheads Wednesday, August 17, 11

  30. Scaling problem: switches • Inherent overheads • Implementation-imposed overheads - Flow setup - Statistics gathering - State size (see paper) Wednesday, August 17, 11

  31. Flow setup OpenFlow controller Client A Client B ProCurve 5406 zl switch Wednesday, August 17, 11

  32. Flow setup OpenFlow controller Client A Client B ProCurve 5406 zl switch We believe our measurement numbers are representative of the current generation of OpenFlow switches Wednesday, August 17, 11

  33. Flow setup 300 275 250 Flow setups per sec. 200 150 100 50 0 5406 zl Wednesday, August 17, 11

  34. Flow setup We can expect up 10,000 10000 to 10K fl ow 9000 arrivals / sec. 8000 [Benson et al. IMC 2010] Flow setups per sec. 7000 40x di ff erence! 6000 5000 4000 3000 2000 1000 0 5406 zl Expected Wednesday, August 17, 11

  35. Flow setup 10,000 10000 9000 Too much latency: 8000 Flow setups per sec. adds 2ms to fl ow 7000 setup 6000 5000 4000 3000 2000 1000 0 5406 zl Expected Wednesday, August 17, 11

  36. Control-plane 300 Gbps Switch CPU Inbound Routed packets packets Data-plane Wednesday, August 17, 11

  37. Control-plane 80 Mbps Switch CPU Inbound Routed packets packets Data-plane Wednesday, August 17, 11

  38. Control-plane 17 Mbps Switch CPU Inbound Routed packets packets Data-plane Wednesday, August 17, 11

  39. Stats-gathering • Flow setups and stat-pulling compete for this bandwidth Wednesday, August 17, 11

  40. Stats-gathering • Flow setups and stat-pulling compete for this bandwidth 300 Flow setups per sec. 250 200 150 100 50 0 Stat-pull frequency: Never 1 s 500 ms Wednesday, August 17, 11

  41. Stats-gathering • Flow setups and stat-pulling compete for this bandwidth - 2.5 sec. to collect stats from the average data center edge switch Wednesday, August 17, 11

  42. Can we solve the problem with more hardware? • Faster CPU may help, but won’t be enough - Control-plane datapath needs at least two orders of magnitude more bandwidth • Ethernet speeds accelerating faster than CPU speeds • OpenFlow won’t drive chip-area budgets for several generations Wednesday, August 17, 11

  43. Contributions • Characterize overheads of implementing OpenFlow in hardware • Propose DevoFlow to enable cost- e ff ective, scalable fl ow management • Evaluate DevoFlow by applying it to data center fl ow scheduling Wednesday, August 17, 11

  44. Devolved OpenFlow We devolve control over most fl ows back to the switches Wednesday, August 17, 11

  45. DevoFlow design • Keep fl ows in the data-plane • Maintain just enough visibility for e ff ective fl ow management • Simplify the design and implementation of high-performance switches Wednesday, August 17, 11

  46. DevoFlow mechanisms • Control mechanisms • Statistics-gathering mechanisms Wednesday, August 17, 11

  47. DevoFlow mechanisms • Control mechanisms - Rule cloning • ASIC clones a wildcard rule as an exact match rule for new micro fl ows Wednesday, August 17, 11

  48. DevoFlow mechanisms • Control mechanisms - Rule cloning src dst src port dst Port wildcard rules * 129.100.1.5 * * src dst src port dst Port exact-match rules Wednesday, August 17, 11

  49. DevoFlow mechanisms • Control mechanisms - Rule cloning src dst src port dst Port wildcard rules * 129.100.1.5 * * src dst src port dst Port exact-match rules Wednesday, August 17, 11

  50. DevoFlow mechanisms • Control mechanisms - Rule cloning src dst src port dst Port wildcard rules * 129.100.1.5 * * src dst src port dst Port exact-match rules Wednesday, August 17, 11

  51. DevoFlow mechanisms • Control mechanisms - Rule cloning src dst src port dst Port wildcard rules * 129.100.1.5 * * src dst src port dst Port exact-match 129.200.1.1 129.100.1.5 4832 80 rules Wednesday, August 17, 11

  52. DevoFlow mechanisms • Control mechanisms - Rule cloning • ASIC clones a wildcard rule as an exact match rule for new micro fl ows - Local actions • Rapid re-routing • Gives fallback paths for when a port fails • Multipath support Wednesday, August 17, 11

  53. Control-mechanisms • Control mechanisms - Rule cloning • ASIC clones a wildcard rule as an exact match rule for new micro fl ows - Local actions • 1/3 1/6 1/2 Rapid re-routing • Gives fallback paths for when a port fails • Multipath support Wednesday, August 17, 11

  54. Statistics-gathering mechanisms Wednesday, August 17, 11

  55. Statistics-gathering mechanisms • Sampling - Packet header is sent to controller with 1/1000 probability Wednesday, August 17, 11

Download Presentation
Download Policy: The content available on the website is offered to you 'AS IS' for your personal information and use only. It cannot be commercialized, licensed, or distributed on other websites without prior consent from the author. To download a presentation, simply click this link. If you encounter any difficulties during the download process, it's possible that the publisher has removed the file from their server.

Recommend

Greenland ice sheet flow computations scaling-up to high spatial resolution and fast boundary

Greenland ice sheet flow computations scaling-up to high spatial resolution and fast boundary

Greenland ice sheet flow computations scaling-up to high spatial resolution and fast boundary processes Ed Bueler Dept. of Mathematics and Statistics Geophysical Institute University of Alaska, Fairbanks with help from Andy Aschwanden, ARSC,

373 views • 35 slides
Performance Scaling How is my parallel code performing and scaling? Performance metrics

Performance Scaling How is my parallel code performing and scaling? Performance metrics

Performance Scaling How is my parallel code performing and scaling? Performance metrics Measure the execution time T how do we quantify performance improvements? Speed up typically S(N,P) < P Parallel efficiency

377 views • 17 slides
Towards High- -performance performance Towards High Flow- -level Packet Processing level

Towards High- -performance performance Towards High Flow- -level Packet Processing level

Towards High- -performance performance Towards High Flow- -level Packet Processing level Packet Processing Flow on Multi- -core Network Processors core Network Processors on Multi Yaxuan Qi (presenter), Bo Xu, Fei He, Baohua Yang,

568 views • 27 slides
Noria Dynamic, partially-stateful data-flow for high-performance Web applications Jon

Noria Dynamic, partially-stateful data-flow for high-performance Web applications Jon

Noria Dynamic, partially-stateful data-flow for high-performance Web applications Jon Gjengset Malte Schwarzkopf Jonathan Behrens Lara Timb Arajo Martin Ek Eddie Kohler M. Frans Kaashoek Robert Morris 2 Frontend 2

1.6k views • 108 slides
Cell Selection for High-Performance Designs in an Industrial Design Flow PROGRAMA DE

Cell Selection for High-Performance Designs in an Industrial Design Flow PROGRAMA DE

Tiago Reimann Cliff Sze Ricardo Reis ISPD - 05 Apr 2016 Cell Selection for High-Performance Designs in an Industrial Design Flow PROGRAMA DE PS-GRADUAO EM MICROELETRNICA Power Battery life Power/Cooling bill 2 Outline

636 views • 28 slides
URBAN FLOW MODELING AND SOLAR FORECASTING USING HIGH-PERFORMANCE COMPUTING Harish Gopalan,

URBAN FLOW MODELING AND SOLAR FORECASTING USING HIGH-PERFORMANCE COMPUTING Harish Gopalan,

URBAN FLOW MODELING AND SOLAR FORECASTING USING HIGH-PERFORMANCE COMPUTING Harish Gopalan, Venugopalan Raghavan , Senthil Kumar Selvaraj, Chin Chun Ooi, Arthur Teck-Bin Lim, George Xu, Pao-Hsiung Chiu, Su Yi, Poh Hee Joo and Lou Jing Fluid

700 views • 36 slides
Getting the Performance Out Of Getting the Performance Out Of High Performance Computing High

Getting the Performance Out Of Getting the Performance Out Of High Performance Computing High

Getting the Performance Out Of Getting the Performance Out Of High Performance Computing High Performance Computing Jack Dongarra I nnovative Computing Lab University of Tennessee and Computer Science and Math Division Oak Ridge Nat ional

380 views • 13 slides
Ecology and Management of Headwater Streams Groundwater Steep and Narrow Channels High flow

Ecology and Management of Headwater Streams Groundwater Steep and Narrow Channels High flow

Ecology and Management of Headwater Streams Groundwater Steep and Narrow Channels High flow (spring and after rain) Base flow (early summer) No flow (late summer/fall) Mapped USGS Streams (1:24,000 scale) 19,553 Acres 21,405 Acres Mapped

622 views • 33 slides
Uniprocessor Performance not Scaling Performance (vs. VAX-11/780) 10000 20%/year Concurrency

Uniprocessor Performance not Scaling Performance (vs. VAX-11/780) 10000 20%/year Concurrency

Uniprocessor Performance not Scaling Performance (vs. VAX-11/780) 10000 20%/year Concurrency 1000 and Threads 100 52%/year 10 25%/year 1 1978 1980 1982 1984 1986 1988 1990 1992 1994 1996 1998 2000 2002 2004 2006 Source: David Patterson

328 views • 12 slides
Building a High Performance Team to Achieve Excellence in Facility Management . June 3,

Building a High Performance Team to Achieve Excellence in Facility Management . June 3,

Building a High Performance Team to Achieve Excellence in Facility Management . June 3, 2015 The Journey Begins High-Performance FM Organizations What does HP/world class mean to you? How do you get there? What role

669 views • 42 slides
Adaptive and Localized Basis Functions for O ( N ) vers. Linear Scaling, Large Systems and

Adaptive and Localized Basis Functions for O ( N ) vers. Linear Scaling, Large Systems and

Max Conference on the Materials Design Ecosystem at the Exascale: High-Performance and High-Throughput Computing T RIESTE bigdft.org Cubic vers. Adaptive and Localized Basis Functions for O ( N ) vers. Linear Scaling, Large Systems and

462 views • 28 slides
Scaling properties in multijet events towards high multiplicities Peter Schichtel

Scaling properties in multijet events towards high multiplicities Peter Schichtel

Scaling properties in multijet events towards high multiplicities Peter Schichtel Durham University @ Higgs plus Jets 2014 Content Introduction theoretical uncertainties in multi-jet observables jet scaling patterns Jet

451 views • 18 slides
ENSC 835- -3: 3: HIGH HIGH- -PERFORMANCE PERFORMANCE ENSC 835 NETWORKS NETWORKS DYNAMIC

ENSC 835- -3: 3: HIGH HIGH- -PERFORMANCE PERFORMANCE ENSC 835 NETWORKS NETWORKS DYNAMIC

ENSC 835- -3: 3: HIGH HIGH- -PERFORMANCE PERFORMANCE ENSC 835 NETWORKS NETWORKS DYNAMIC RIGHT- -SIZING SIZING DYNAMIC RIGHT A TCP FLOW- -CONTROL INVESTIGATION IN CONTROL INVESTIGATION IN A TCP FLOW OPNET OPNET by by CAMILLE

1.01k views • 21 slides
So#ware Scaling Mo/va/on & Goals HW Configura/on & Scale Out So#ware Scaling

So#ware Scaling Mo/va/on & Goals HW Configura/on & Scale Out So#ware Scaling

So#ware Scaling Mo/va/on & Goals HW Configura/on & Scale Out So#ware Scaling Efforts System management Opera/ng system Programming environment PreAcceptance Work HW stabiliza/on & early scaling

390 views • 19 slides
High Flow Experiment Assessment Summary Adaptive Management Workgroup Webinar May 22, 2019

High Flow Experiment Assessment Summary Adaptive Management Workgroup Webinar May 22, 2019

High Flow Experiment Assessment Summary Adaptive Management Workgroup Webinar May 22, 2019 Scott VanderKooi Grand Canyon Monitoring and Research Center Southwest Biological Science Center U.S. Department of the Interior U.S. Geological

718 views • 35 slides
Lower Yuba River Accord Accord Flow and Temperature Performance Yuba Accord River Management Team

Lower Yuba River Accord Accord Flow and Temperature Performance Yuba Accord River Management Team

Lower Yuba River Accord Accord Flow and Temperature Performance Yuba Accord River Management Team 5 th Annual Symposium June 12, 2013 Accord Flow and Temperature Overview Flows Temperatures Implementation of Yuba Accord Operations

317 views • 20 slides
Chapter 6 Conclusions Summary and Outlook NIKOLAOS KONSTANTINOU DIMITRIOS-EMMANUEL SPANOS

Chapter 6 Conclusions Summary and Outlook NIKOLAOS KONSTANTINOU DIMITRIOS-EMMANUEL SPANOS

Chapter 6 Conclusions Summary and Outlook NIKOLAOS KONSTANTINOU DIMITRIOS-EMMANUEL SPANOS Materializing the Web of Linked Data Outline Introduction Recap Discussion Open Research Challenges Chapter 6 Materializing the Web of Linked Data

618 views • 31 slides
Visualise a web site with tag clouds generated by R Sigbert Klinke 1 , 2 1 Institute for Statistics

Visualise a web site with tag clouds generated by R Sigbert Klinke 1 , 2 1 Institute for Statistics

Introduction Visualise a web site with tag clouds generated by R Sigbert Klinke 1 , 2 1 Institute for Statistics and Econometrics, School of Business and Economics, Humboldt-Universit at zu Berlin 2 Business and Human Resource Education, Dept.

346 views • 13 slides
Active BGP Measurement with BGP-Mux Ethan Katz-Bassett (USC) with testbed and some slides

Active BGP Measurement with BGP-Mux Ethan Katz-Bassett (USC) with testbed and some slides

Active BGP Measurement with BGP-Mux Ethan Katz-Bassett (USC) with testbed and some slides hijacked from Nick Feamster and Valas Valancius 2 Before I Start Georgia Tech system, I am just an enthusiastic user Nick Feamster and his

874 views • 62 slides
The SLAC ATCA Platforms Ryan Herbst, SLAC National Accelerator Laboratory Current ATCA

The SLAC ATCA Platforms Ryan Herbst, SLAC National Accelerator Laboratory Current ATCA

The SLAC ATCA Platforms Ryan Herbst, SLAC National Accelerator Laboratory Current ATCA Development SLAC has been focused on ATCA based data acquisition and control systems RCE (Reconfigurable Cluster Element) platform is a full meshed

731 views • 28 slides
SAP Business One for Warehouse Management (WMS) So Solution Overvie iew Global VAR for SAP

SAP Business One for Warehouse Management (WMS) So Solution Overvie iew Global VAR for SAP

SAP Business One for Warehouse Management (WMS) So Solution Overvie iew Global VAR for SAP Business One Solution Benefits Compliance with industry standards Inbound logistics Outbound logistics Supports GS1/UCC standards

920 views • 15 slides
Network Visualization Introduction Presented by Shahed Introduction Introduction Basic

Network Visualization Introduction Presented by Shahed Introduction Introduction Basic

Network Visualization Introduction Presented by Shahed Introduction Introduction Basic building blocks Node Links (relationship between nodes) Spatial information Network data

584 views • 9 slides
$Q(PSLULFDO6WXG\RI5HDO ,QWURGXFWLRQ $XGLR7UDIILF ,QWHUQHWLVJURZLQJ

$Q(PSLULFDO6WXG\RI5HDO ,QWURGXFWLRQ $XGLR7UDIILF ,QWHUQHWLVJURZLQJ

$Q(PSLULFDO6WXG\RI5HDO ,QWURGXFWLRQ $XGLR7UDIILF ,QWHUQHWLVJURZLQJ :HEIDFLOLWDWHVLQWHJUDWLRQRIVWUHDPLQJDXGLR 5DGLRMXNHER[HV $ 0HQD DQG-+HLGHPDQQ %URDGFDVWUDGLR

233 views • 7 slides
URLRewri)ngforGood,notEvil UsingAlterna)veResourceLocators

URLRewri)ngforGood,notEvil UsingAlterna)veResourceLocators

URLRewri)ngforGood,notEvil UsingAlterna)veResourceLocators BryanSullivan SeniorSecurityProgramManager,SDL MicrosoA TopWebVulnsHaveaCommon Factor CrossSiteScrip)ng OWASP#1

640 views • 20 slides

More recommend