fab towards flow aware bu fg er sharing in programmable
play

FAB: Towards Flow-aware Bu fg er Sharing in Programmable Switches - PowerPoint PPT Presentation

Mar 02, 2024 •183 likes •1.04k views

FAB: Towards Flow-aware Bu fg er Sharing in Programmable Switches Maria Apostolaki Joint work with Laurent Vanbever & Manya Ghobadi An old story An old story Fan-in causing queue built-up Switch Senders Receivers An

  1. FAB: Towards Flow-aware Bu fg er Sharing in Programmable Switches Maria Apostolaki Joint work with Laurent Vanbever & Manya Ghobadi

  2. An old story…

  3. An old story… Fan-in causing queue built-up … Switch … Senders Receivers

  4. An old story… Fan-in causing drops Drops … Switch … Senders Receivers

  5. An old story… Drops increase Flow Completion Times Drops … Switch … High SLA Drops tail FCT violations

  6. An old story… Multiple approaches to address this issue Active Queue Management Drops … TCP Version Switch Scheduling …

  7. An old story… Multiple approaches to address this issue Active Queue Management Drops … TCP Version Switch Scheduling …

  8. An old story… Multiple approaches to address this issue Active Queue Management Drops … TCP Version Switch Scheduling …

  9. An old story… Multiple approaches to address this issue Active Queue Management Drops … TCP Version Switch Scheduling …

  10. Bu fg er management : the algorithm according to which ports/queues of a device share a common bu fg er

  11. Bu fg er management : the algorithm according to which ports/queues of a device share a common bu fg er Most of today’s devices have a shared bu fg er to absorb bursts

  12. How many packets can each port store in the common bu fg er? … Switch …

  13. How many packets can each port store in the common bu fg er? remaining excessive packets will be dropped … Switch …

  14. How many packets can each port store in the common bu fg er? remaining excessive packets will be dropped … Switch Let’s give… … …half of the bu fg er to each port! …small fraction to each port!

  15. FAB: Towards Flow-aware Bu fg er Sharing in Programmable Switches Joint work with Laurent Vanbever & Manya Ghobadi

  16. Outline Background FAB Initial Results Practicality

  17. Outline Background FAB Initial Results Practicality

  18. Three common bu fg er management techniques with pros & cons Complete Partitioning: works for balanced tra ffj c wastes bu fg er otherwise statically allocated bu fg er space per port Complete Sharing: works for unbalanced tra ffj c benefits long flows unrestricted bu fg er space per port Dynamic Sharing: adapts fairly to load fraction ( α ) of remaining bu fg er per port ignores queue content

  19. Three common bu fg er management techniques with pros & cons Complete Partitioning: works for balanced tra ffj c wastes bu fg er otherwise statically allocated bu fg er space per port Complete Sharing: works for unbalanced tra ffj c benefits long flows unrestricted bu fg er space per port Dynamic Sharing: adapts fairly to load fraction ( α ) of remaining bu fg er per port ignores queue content

  20. Three common bu fg er management techniques with pros & cons Complete Partitioning: works for balanced tra ffj c wastes bu fg er otherwise statically allocated bu fg er space per port Complete Sharing: works for unbalanced tra ffj c benefits long flows unrestricted bu fg er space per port Dynamic Sharing: adapts fairly to load fraction ( α ) of remaining bu fg er per port ignores queue content

  21. Three common bu fg er management techniques with pros & cons Complete Partitioning: works for balanced tra ffj c wastes bu fg er otherwise statically allocated bu fg er space per port Complete Sharing: works for unbalanced tra ffj c benefits long flows unrestricted bu fg er space per port Dynamic Sharing: adapts fairly to load fraction ( α ) of remaining bu fg er per port ignores queue content

  22. Back to our story… Simulation Results … Switch … Senders Receivers

  23. Simulation Results Shared bu fg er can host up to 180 packets 180 packets … Switch … Senders Receivers

  24. Simulation Results Long flows will consume as much bu fg er as there is available 180 Long Flows # pkts bu fg ered 135 90 … 45 port 1 0 Switch 0 0.4 0.8 1.2 Time (sec) … Senders Receivers

  25. Short flows will attempt to instantaneously store Simulation Results at most 75 packets in the shared bu fg er 180 Long Flows # pkts bu fg ered 135 90 … 45 port 1 0 Switch 0 0.4 0.8 1.2 Time (sec) port 2 Short Flows 80 # pkts bu fg ered 60 40 … 20 0 0 0.4 0.8 1.2 Time (sec) Senders Receivers

  26. Three common bu fg er management techniques with pros & cons Complete Partitioning: works for balanced tra ffj c wastes bu fg er otherwise statically allocated bu fg er space per port Complete Sharing: works for unbalanced tra ffj c benefits long flows unrestricted bu fg er space per port Dynamic Sharing: adapts fairly to load fraction ( α ) of remaining bu fg er per port ignores queue content

  27. Complete Partitioning: static bu fg er space per port = 10 packets 100 100 80 80 # Pkts in Buffer # Pkts in Bu � er # Pkts in Bu � er 60 60 40 40 20 20 0 0 2x10 8 2x10 8 4x10 8 4x10 8 6x10 8 6x10 8 8x10 8 8x10 8 1x10 9 1x10 9 1.2x10 9 1.2x10 9 0.2 0.4 0.6 0.6 0.8 1 Time (s) 0 0 Time (ns) Time (ns)

  28. Receiver of long flows bu fg ers up to 10 packets 100 100 80 80 # Pkts in Buffer # Pkts in Bu � er # Pkts in Bu � er 60 60 40 40 20 20 0 0 2x10 8 2x10 8 4x10 8 4x10 8 6x10 8 6x10 8 8x10 8 8x10 8 1x10 9 1x10 9 1.2x10 9 1.2x10 9 0.2 0.4 0.6 0.6 0.8 1 Time (s) 0 0 Time (ns) Time (ns)

  29. Receiver of long flows bu fg ers up to 10 packets 100 80 # Pkts in Buffer # Pkts in Bu � er 60 40 20 0 2x10 8 4x10 8 6x10 8 8x10 8 1x10 9 1.2x10 9 0.2 0.4 0.6 0.8 1 Time (s) 0 Time (ns)

  30. Receiver of long flows bu fg ers up to 10 packets used by 5% port 1 100 Unused 95% 80 # Pkts in Buffer # Pkts in Bu � er 60 40 20 0 2x10 8 4x10 8 6x10 8 8x10 8 1x10 9 1.2x10 9 0.2 0.4 0.6 0.8 1 Time (s) 0 Time (ns)

  31. Receiver of short flows bu fg ers up to 10 packets 100 80 # Pkts in Buffer # Pkts in Bu � er 60 40 20 0 2x10 8 4x10 8 6x10 8 8x10 8 1x10 9 1.2x10 9 0.2 0.4 0.6 0.8 1 Time (s) 0 Time (ns)

  32. Receiver of short flows bu fg ers up to 10 packets 100 100 80 80 # Pkts in Buffer # Pkts in Bu � er # Pkts in Bu � er 60 60 40 40 20 20 0 0 2x10 8 2x10 8 4x10 8 4x10 8 6x10 8 6x10 8 8x10 8 8x10 8 1x10 9 1x10 9 1.2x10 9 1.2x10 9 0.2 0.4 0.6 0.8 1 Time (s) 0 0 Time (ns) Time (ns)

  33. Bu fg er is 90% empty used by 5% 5% port 2 100 100 Unused 90% 80 80 # Pkts in Buffer # Pkts in Bu � er # Pkts in Bu � er 60 60 40 40 20 20 0 0 2x10 8 2x10 8 4x10 8 4x10 8 6x10 8 6x10 8 8x10 8 8x10 8 1x10 9 1x10 9 1.2x10 9 1.2x10 9 0.2 0.4 0.6 0.8 1 Time (s) 0 0 Time (ns) Time (ns)

  34. Bu fg er is 90% empty, yet the burst is not absorbed 5% 5% 100 90% 80 Burst Size # Pkts in Buffer # Pkts in Bu � er 60 40 20 0 2x10 8 4x10 8 6x10 8 8x10 8 1x10 9 1.2x10 9 0.2 0.4 0.6 0.8 1 Time (s) 0 Time (ns)

  35. Three common bu fg er management techniques with pros & cons Complete Partitioning: works for balanced tra ffj c wastes bu fg er otherwise statically allocated bu fg er space per port Complete Sharing: works for unbalanced tra ffj c benefits long flows unrestricted bu fg er space per port Dynamic Sharing: adapts fairly to load fraction ( α ) of remaining bu fg er per port ignores queue content

  36. Three common bu fg er management techniques with pros & cons Complete Partitioning: works for balanced tra ffj c wastes bu fg er otherwise statically allocated bu fg er space per port Complete Sharing: works for unbalanced tra ffj c benefits long flows unrestricted bu fg er space per port Dynamic Sharing: adapts fairly to load fraction ( α ) of remaining bu fg er per port ignores queue content

  37. Complete Sharing: unrestricted bu fg er space per port 200 180 160 # Pkts in Buffer 140 # Pkts in Bu � er 120 100 80 60 40 20 0 2x10 8 0.2 4x10 8 0.4 6x10 8 0.6 8x10 8 0.8 1x10 9 1 1.2x10 9 Time (s) 0 Time (ns)

  38. Long flows use all bu fg er 200 180 160 # Pkts in Buffer 140 # Pkts in Bu � er 120 100 80 60 40 20 0 2x10 8 0.2 4x10 8 0.4 6x10 8 0.6 8x10 8 0.8 1x10 9 1 1.2x10 9 Time (s) 0 Time (ns)

  39. Long flows use all bu fg er 200 200 180 180 160 160 # Pkts in Buffer 140 140 # Pkts in Bu � er # Pkts in Bu � er 120 120 100 100 80 80 60 60 40 40 20 20 0 0 2x10 8 2x10 8 0.2 4x10 8 4x10 8 0.4 6x10 8 6x10 8 0.6 8x10 8 8x10 8 0.8 1x10 9 1x10 9 1 1.2x10 9 1.2x10 9 Time (s) 0 0 Time (ns) Time (ns)

  40. Long flows use all bu fg er 200 200 180 180 100% 160 160 # Pkts in Buffer 140 140 # Pkts in Bu � er # Pkts in Bu � er 120 120 100 100 80 80 60 60 40 40 20 20 0 0 2x10 8 2x10 8 0.2 4x10 8 4x10 8 0.4 6x10 8 6x10 8 0.6 8x10 8 8x10 8 0.8 1x10 9 1x10 9 1 1.2x10 9 1.2x10 9 Time (s) 0 0 Time (ns) Time (ns)

  41. Upon arrival, the burst finds the bu fg er fully occupied 200 200 180 180 160 160 # Pkts in Buffer 140 140 # Pkts in Bu � er # Pkts in Bu � er 120 120 100 100 80 80 60 60 40 40 20 20 0 0 2x10 8 2x10 8 0.2 4x10 8 4x10 8 0.4 6x10 8 6x10 8 0.6 8x10 8 8x10 8 0.8 1x10 9 1x10 9 1 1.2x10 9 1.2x10 9 Time (s) 0 0 Time (ns) Time (ns)

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

MP-HULA A Multipath Transport Layer Aware Datacenter Load Balancing Scheme Using Programmable

MP-HULA A Multipath Transport Layer Aware Datacenter Load Balancing Scheme Using Programmable

MP-HULA A Multipath Transport Layer Aware Datacenter Load Balancing Scheme Using Programmable Data Planes Cristian Hernandez Benet , Andreas J. Kassler, Theophilus Benson, Gergely Pongracz MP-HULA: A Transport Layer aware Load Balancing

541 views • 43 slides
APPLICATION-AWARE FLOW MONITORING Thursday 11 th April, 2019 Petr Velan Motivation

APPLICATION-AWARE FLOW MONITORING Thursday 11 th April, 2019 Petr Velan Motivation

APPLICATION-AWARE FLOW MONITORING Thursday 11 th April, 2019 Petr Velan Motivation Application-Aware Flow Monitoring Page 2 / 22 Internet Basic Flow Monitoring TAP Probe Flow monitoring is widely used for: Collector Accounting Router

367 views • 23 slides
IIT Bombay CDEEP Autumn 2009 Introduction to Programmable Logic Design Flow Presented by-

IIT Bombay CDEEP Autumn 2009 Introduction to Programmable Logic Design Flow Presented by-

IIT Bombay CDEEP Autumn 2009 Introduction to Programmable Logic Design Flow Presented by- Anil Powai Labs Tech. Pvt. Ltd. EE705/707 Lectures No. 23 and 24 Prof. M. Shojaei Baghini IIT Bombay Programmable Logic

696 views • 48 slides
On Flow-Aware CSMA in Multi-Channel Wireless Networks Mathieu Feuillet Joint work with Thomas

On Flow-Aware CSMA in Multi-Channel Wireless Networks Mathieu Feuillet Joint work with Thomas

On Flow-Aware CSMA in Multi-Channel Wireless Networks Mathieu Feuillet Joint work with Thomas Bonald CISS 2011 Outline Model Background Standard CSMA Flow-aware CSMA Conclusion Outline Model Background Standard CSMA Flow-aware CSMA

903 views • 58 slides
On Flow Concurrency in the Internet and its Implications for Capacity Sharing Brian Trammell and

On Flow Concurrency in the Internet and its Implications for Capacity Sharing Brian Trammell and

On Flow Concurrency in the Internet and its Implications for Capacity Sharing Brian Trammell and Dominik Schatzmann Communications Systems Group, ETH Zrich Whats a measurement guy doing at a capacity-sharing workshop? Concern about

599 views • 13 slides
Resource-Aware Protocols for Network Cost-Sharing Games Moh a m a d L a ti f i a n, Sh a rif

Resource-Aware Protocols for Network Cost-Sharing Games Moh a m a d L a ti f i a n, Sh a rif

Resource-Aware Protocols for Network Cost-Sharing Games Moh a m a d L a ti f i a n, Sh a rif University of Technology George Christodoulou , University of Liverpool Vasilis Gkatzelis , Drexel University Alkmini Sgouritsa , University of Liverpool

1k views • 88 slides
OptFlow: A Flow-based Abstraction for Programmable Topologies Klaus-Tycho Foerster Long Luo

OptFlow: A Flow-based Abstraction for Programmable Topologies Klaus-Tycho Foerster Long Luo

OptFlow: A Flow-based Abstraction for Programmable Topologies Klaus-Tycho Foerster Long Luo Manya Ghobadi Computer Science and Artificial Faculty of Computer Science University of Electronic Science University of Vienna and Technology of

894 views • 33 slides
QoS-Aware Adaptive Flow-Rule Aggregation in Software-Defined IoT N. Saha, S. Misra and S. Bera

QoS-Aware Adaptive Flow-Rule Aggregation in Software-Defined IoT N. Saha, S. Misra and S. Bera

QoS-Aware Adaptive Flow-Rule Aggregation in Software-Defined IoT N. Saha, S. Misra and S. Bera Department of Computer Science and Engineering, Indian Institute of Technology, Kharagpur, India IEEE GLOBECOM 2018, Abu Dhabi, UAE Problem

219 views • 10 slides
Loss-Constrained Minimum Cost Flow under Arc Failure Uncertainty with Applications in Risk-Aware

Loss-Constrained Minimum Cost Flow under Arc Failure Uncertainty with Applications in Risk-Aware

Loss-Constrained Minimum Cost Flow under Arc Failure Uncertainty with Applications in Risk-Aware Kidney Exchange Siqian Shen University of Michigan at Ann Arbor joint work with Qipeng Zheng & Yuhui Shi (Univ. of Central Florida; Univ. of

753 views • 37 slides
Towards a Reliability-aware Design Flow for Kahn Process Networks on NoC-based Multiprocessors

Towards a Reliability-aware Design Flow for Kahn Process Networks on NoC-based Multiprocessors

Towards a Reliability-aware Design Flow for Kahn Process Networks on NoC-based Multiprocessors Onur Derin , Leandro Fiorin ALaRI, Faculty of Informatics, University of Lugano, Switzerland { derino,fiorin } @alari.ch L ubeck Feb 25, 2014

415 views • 27 slides
Field Programmable Gate Arrays by Ketil Red Field Programmable Gate Array Integrated

Field Programmable Gate Arrays by Ketil Red Field Programmable Gate Array Integrated

A brief introduction to Field Programmable Gate Arrays by Ketil Red Field Programmable Gate Array Integrated circuit including a matrix of general-purpose programmable logic I I I I I I I I I I I I I I I I I I I I

439 views • 14 slides
PROGRAMMABLE LOGIC CONTROLLER Control Systems Types Programmable Logic Controllers

PROGRAMMABLE LOGIC CONTROLLER Control Systems Types Programmable Logic Controllers

PROGRAMMABLE LOGIC CONTROLLER Control Systems Types Programmable Logic Controllers Distributed Control System PC- Based Controls Programmable Logic Controllers PLC Sequential logic solver PID Calculations. Advanced

292 views • 27 slides
Regulatory Guidance on the Use of Field Programmable Gate of Field Programmable Gate Arrays in

Regulatory Guidance on the Use of Field Programmable Gate of Field Programmable Gate Arrays in

Regulatory Guidance on the Use of Field Programmable Gate of Field Programmable Gate Arrays in the U.S. October 13, 2015 Steven A. Arndt, Ph.D., P.E. Office of Nuclear Reactor Regulation The views expressed in this presentation are solely

542 views • 23 slides
ROMs, PLAs and FPGAs October 5, 2006 Typeset by Foil T EX Why Programmable Logic?

ROMs, PLAs and FPGAs October 5, 2006 Typeset by Foil T EX Why Programmable Logic?

ROMs, PLAs and FPGAs October 5, 2006 Typeset by Foil T EX Why Programmable Logic? Programmable logic technologies: Read-Only Memory (ROM) Programmable Logic Array (PLA) Programmable Array Logic (PAL) Field Programmable

283 views • 13 slides
Stress-Aware Routing to Mitigate Aging Effects in SRAM-based FPGAs International

Stress-Aware Routing to Mitigate Aging Effects in SRAM-based FPGAs International

Stress-Aware Routing to Mitigate Aging Effects in SRAM-based FPGAs International Conference on Field- Programmable Logic and Applications Lausanne, Switzerland, Aug. 29 th Sep. 2 nd , 2016 Behnam Khaleghi , Behzad Omidi

539 views • 23 slides
Programmable Switch Hardware ECE/CS598HPN Radhika Mittal Conventional SDN Programmable

Programmable Switch Hardware ECE/CS598HPN Radhika Mittal Conventional SDN Programmable

Programmable Switch Hardware ECE/CS598HPN Radhika Mittal Conventional SDN Programmable control plane . Data plane can support high bandwidth. But has limited flexibility. Restricted to conventional packet protocols. Software

556 views • 40 slides
Dataflow Network Programming with Neuron Eric Griffis RacketCon 2018 St. Louis, MO 1 I. Meet

Dataflow Network Programming with Neuron Eric Griffis RacketCon 2018 St. Louis, MO 1 I. Meet

Dataflow Network Programming with Neuron Eric Griffis RacketCon 2018 St. Louis, MO 1 I. Meet Nick 2 Nick cares about cancer 3 Curing cancer is expensive 4 5 6 II. Enter Neuron 7 The Neuron Framework concurrency model messaging API

580 views • 28 slides
Cap Sante North & West Basin Upland Redevelopment Council Briefjng - February 24, 2020

Cap Sante North & West Basin Upland Redevelopment Council Briefjng - February 24, 2020

Cap Sante North & West Basin Upland Redevelopment Council Briefjng - February 24, 2020 Overview Project Process and Purpose Development Plan Site Plan Rationale Event Center Building Design Parking Strategy

840 views • 61 slides
Consistent Reads Using ProxySQL and GTID Santa Clara, California | April 23th 25th, 2018

Consistent Reads Using ProxySQL and GTID Santa Clara, California | April 23th 25th, 2018

Consistent Reads Using ProxySQL and GTID Santa Clara, California | April 23th 25th, 2018 Disclaimer I am not Ren Canna @lefred MySQL Community Manager / Oracle the one who provided a hint for this not the one

1.27k views • 50 slides
Reserved: Dissectjng Internet Traffjc on Port 0 Aniss Maghsoudlou Oliver Gasser Anja Feldmann

Reserved: Dissectjng Internet Traffjc on Port 0 Aniss Maghsoudlou Oliver Gasser Anja Feldmann

Reserved: Dissectjng Internet Traffjc on Port 0 Aniss Maghsoudlou Oliver Gasser Anja Feldmann Max Planck Instjtute for Informatjcs Why Port 0? Using port number 0 is not allowed in: TCP [RFC 1340] UDP [RFC 8085] 76 GB of traffjc

234 views • 10 slides
Covert channels in TCP/IP: attack and defence The creation and detection of TCP/IP steganography

Covert channels in TCP/IP: attack and defence The creation and detection of TCP/IP steganography

Covert channels in TCP/IP: attack and defence The creation and detection of TCP/IP steganography for covert channels and device fingerprinting Steven J. Murdoch and Stephen Lewis http://www.cl.cam.ac.uk/users/ { sjm217 , srl32 } Computer

915 views • 36 slides
NFS Don Porter CSE 506 together as the !handle for a rue. The inode generation number is

NFS Don Porter CSE 506 together as the !handle for a rue. The inode generation number is

NFS Don Porter CSE 506 together as the !handle for a rue. The inode generation number is necessary because the server may hand out an !handle with an inode number of a file that is later removed and the inode J reused. When the original

617 views • 24 slides
Information Management for Business Dr. Tom Acton 1 2 3 Huh? Whats an organisation?

Information Management for Business Dr. Tom Acton 1 2 3 Huh? Whats an organisation?

Information Management for Business Dr. Tom Acton 1 2 3 Huh? Whats an organisation? What do businesses do anyway? What are the parts of a business, and what has technology got to do with any of it? And hold on , whats

197 views • 16 slides
Management Nima Honarmand (Based on slides by Don Porter and Mike Ferdman) Fall 2014 :: CSE 506

Management Nima Honarmand (Based on slides by Don Porter and Mike Ferdman) Fall 2014 :: CSE 506

Fall 2014 :: CSE 506 :: Section 2 (PhD) Page Frame Management Nima Honarmand (Based on slides by Don Porter and Mike Ferdman) Fall 2014 :: CSE 506 :: Section 2 (PhD) Recap and background Page tables: translate virtual addresses to physical

366 views • 24 slides

More recommend