size based queuing
play

Size-Based Queuing An Approach to Improve Bandwidth Utilization in - PowerPoint PPT Presentation

Size-Based Queuing An Approach to Improve Bandwidth Utilization in TSN Networks Florian Heilmann and Gerhard Fohler RTN18 - Barcelona, Spain In this talk Highest-priority (HP) messages may experience jitter due to blocking by


  1. Size-Based Queuing An Approach to Improve Bandwidth Utilization in TSN Networks Florian Heilmann and Gerhard Fohler RTN’18 - Barcelona, Spain

  2. In this talk • Highest-priority (HP) messages may experience jitter due to blocking by lower-priority (LP) messages • In a TAS-enabled non-preemptive TSN network • Guard bands • used to avoid this jitter • But also decrease bandwidth utilization • Our approach to reclaim wasted bandwidth • Provide an initial analysis of our approach Size-Based Queuing - An Approach to Improve Bandwidth 2 Utilization in TSN Networks

  3. TSN and TAS primer Size-Based Queuing - An Approach to Improve Bandwidth 3 Utilization in TSN Networks

  4. Ethernet and Time-Sensitive Networking • Switched Ethernet is replacing existing network technologies • Cheap, mature technology • However: No timing/delivery guarantees Size-Based Queuing - An Approach to Improve Bandwidth 4 Utilization in TSN Networks

  5. Ethernet and Time-Sensitive Networking • Continuation of AVB TG • Goal is to make Ethernet safety-critical • Automotive, Avionics, Factory automation • Extend AVB, e.g. • higher precision clock sync • Frame preemption • Frame replication for redundancy Size-Based Queuing - An Approach to Improve Bandwidth 5 Utilization in TSN Networks

  6. Time-Aware Shaper and Guard Bands • Time-Aware Shaper (TAS) is part of TSN • TAS adds … • Gating mechanism for switch queues • Gate schedule to control the gates • TAS enables … • Scheduled transmissions • Ways to provide exclusive network access to a queue Size-Based Queuing - An Approach to Improve Bandwidth 6 Utilization in TSN Networks

  7. Time-Aware Shaper and Guard Bands 802.1Q Ethernet Switch Safety-critical, highest-priority traffic 𝐻 7 𝑅 7 (denoted HP in this presentation) Ingress Port 1 𝐻 6 𝑅 6 Input queuing function Safety-critical, lower-priority traffic TX selection function (denoted LP in this presentation) Ingress 𝐻 5 𝑅 5 Port 2 Egress Port 𝐻 4 𝑅 4 … Best Effort traffic … … (not considered in this presentation) 𝐻 0 𝑅 0 Ingress Port n Size-Based Queuing - An Approach to Improve Bandwidth 7 Utilization in TSN Networks

  8. Time-Aware Shaper and Guard Bands o 𝑯 𝑰𝑸 c 𝑢 o 𝑯 𝑴𝑸 c 𝑢 𝒑𝒗𝒖 𝐻 𝐼𝑄 𝐻 𝑀𝑄 𝑢 “Closed “Closed” “Open” “Open” ” Size-Based Queuing - An Approach to Improve Bandwidth 8 Utilization in TSN Networks

  9. Time-Aware Shaper and Guard Bands • Highest priority (HP) messages may experience blocking by lower priority (LP) messages Unless: • Network provides message preemption functionality • Not considered in this work • Gate schedules are modified to avoid blocking • Guard bands (GB): Preemptively close gate of LP messages to ensure idle network once HP gate opens Size-Based Queuing - An Approach to Improve Bandwidth 9 Utilization in TSN Networks

  10. Time-Aware Shaper and Guard Bands o 𝐻 𝐼𝑄 c 𝑢 o 𝐻 𝑀𝑄 c 𝑢 HP Jitter 𝑝𝑣𝑢 𝑢 Size-Based Queuing - An Approach to Improve Bandwidth 10 Utilization in TSN Networks

  11. Time-Aware Shaper and Guard Bands o 𝐻 𝐼𝑄 c 𝑢 o 𝐻 𝑀𝑄 Guardband c 𝑢 max 𝑝𝑣𝑢 𝑢 Size-Based Queuing - An Approach to Improve Bandwidth 11 Utilization in TSN Networks

  12. Time-Aware Shaper and Guard Bands o 𝐻 𝐼𝑄 c 𝑢 o 𝐻 𝑀𝑄 Guardband c 𝑢 𝑝𝑣𝑢 𝑢 Size-Based Queuing - An Approach to Improve Bandwidth 12 Utilization in TSN Networks

  13. Problem description Size-Based Queuing - An Approach to Improve Bandwidth 13 Utilization in TSN Networks

  14. Bandwidth Wastage caused by Guard Bands • Guard bands cause time intervals where all gates are closed • Pending messages can’t be transmitted. o 𝐻 𝐼𝑄 c • Even though output link is idle 𝑢  Output bandwidth utilization decreases o 𝐻 𝑀𝑄 (bandwidth is wasted) Guardband c 𝑢 𝑝𝑣𝑢 𝑢 Size-Based Queuing - An Approach to Improve Bandwidth 14 Utilization in TSN Networks

  15. Approaches to reduce Jitter of HP messages • Guard Bands • Can result in reduced bandwidth utilization • Because we always need to consider largest message • Our approach: Size Based Queuing (SBQ) • Builds on the guard band approach Size-Based Queuing - An Approach to Improve Bandwidth 15 Utilization in TSN Networks

  16. Our approach: Size-Based Queuing Size-Based Queuing - An Approach to Improve Bandwidth 16 Utilization in TSN Networks

  17. Size-Based Queuing • Non-preemptive network + Low Jitter requirement of HP messages  guard bands unavoidable • Can we modify existing guard bands? Idea • Guard band size  Size of largest message in queue • Distribute messages into multiple queues according to their size • Control the size of the largest message in each queue  Shrink guard bands  Improve bandwidth utilization Size-Based Queuing - An Approach to Improve Bandwidth 17 Utilization in TSN Networks

  18. Size-Based Queuing • Method overview 1. Split some/all LP queues into queue-sets 2. Define size-thresholds 3. Modify 𝑔 𝑗𝑜 to distribute messages according to thresholds 4. Add new gates for the new queues 5. Update gate schedule or modify 𝑔 𝑝𝑣𝑢 to handle new queues 6. Shrink guard bands where applicable Size-Based Queuing - An Approach to Improve Bandwidth 18 Utilization in TSN Networks

  19. SBQ Implementation Example TSN TAS Ethernet Switch 𝐻 7 𝑅 7 Ingress Port 1 𝐻 6 𝑅 6 Input queuing function TX selection function Ingress Port 2 Egress 𝐻 5 𝑅 5 Port … 𝐻 4 𝑅 4 … … Ingress Port n 𝐻 0 𝑅 0 Size-Based Queuing - An Approach to Improve Bandwidth 19 Utilization in TSN Networks

  20. SBQ Implementation Example 𝐻 6,𝑚𝑠𝑕 𝑅 6,𝑚𝑠𝑕 𝜐 6 𝐻 6 𝐻 6 𝑅 6 𝑅 6 Input queuing function 𝐻 6,𝑡𝑛 𝑅 6,𝑡𝑛 TX selection function 𝐻 5,𝑚𝑠𝑕 𝑅 5,𝑚𝑠𝑕 𝜐 5 𝐻 5 𝐻 5 𝑅 5 𝑅 5 𝐻 5,𝑡𝑛 𝑅 5,𝑡𝑛

  21. SBQ Implementation Example SBQ Ethernet Switch 𝐻 7 𝑅 7 Ingress Port 1 𝐻 6,𝑚𝑠𝑕 𝑅 6,𝑚𝑠𝑕 𝜐 6 Input queuing function 𝐻 6,𝑡𝑛 𝑅 6,𝑡𝑛 TX selection function Ingress Port 2 𝐻 5,𝑚𝑠𝑕 𝑅 5,𝑚𝑠𝑕 Egress 𝜐 5 Port 𝐻 5,𝑡𝑛 𝑅 5,𝑡𝑛 … 𝐻 4 𝑅 4 … … Ingress Port n 𝐻 0 𝑅 0 Size-Based Queuing - An Approach to Improve Bandwidth 21 Utilization in TSN Networks

  22. Analysis Size-Based Queuing - An Approach to Improve Bandwidth 22 Utilization in TSN Networks

  23. Analysis - Bandwidth Utilization o 𝐻 𝐼𝑄 c 𝑢 o 𝐻 𝑀𝑄 Guardband c 𝑢 𝑝𝑣𝑢 𝑢 Size-Based Queuing - An Approach to Improve Bandwidth 23 Utilization in TSN Networks

  24. Analysis - Bandwidth Utilization 𝑡𝑗𝑨𝑓 𝑛𝑡𝑕 ≥ 𝜐 𝑀𝑄 → 𝑅 𝑀𝑄,𝑚𝑠𝑕 𝜐 𝑀𝑄 𝑡𝑗𝑨𝑓 𝑛𝑡𝑕 < 𝜐 𝑀𝑄 → 𝑅 𝑀𝑄,𝑡𝑛 Size-Based Queuing - An Approach to Improve Bandwidth 24 Utilization in TSN Networks

  25. Analysis - Bandwidth Utilization 𝐻 𝐼𝑄 o c 𝑢 “Closed” o 𝐻 𝑀𝑄,𝑚𝑠𝑕 Guardband c “Open” 𝑢 o 𝐻 𝑀𝑄,𝑡𝑛 Guardband GB c 𝑢 max 𝑝𝑣𝑢 𝑢 Size-Based Queuing - An Approach to Improve Bandwidth 25 Utilization in TSN Networks

  26. Analysis – Design Time Overheads • Queues added by SBQ have to be considered during network design • Generate schedules that provide service to all queues • Make use of TSN shapers (CBS, BLS, etc.) • Use additional algorithms to arbitrate within queue-sets • E.g. Round-Robin Size-Based Queuing - An Approach to Improve Bandwidth 26 Utilization in TSN Networks

  27. Analysis - Latency • Changing the way messages are queued affects their latency through a switch • Only LP messages affected • HP unaffected • (we don’t care about BE) Size-Based Queuing - An Approach to Improve Bandwidth 27 Utilization in TSN Networks

  28. Analysis - Latency • Individual queues of a queue set get only a portion of the bandwidth  Time before message reaches front of queue is longer  Latency increases . 1 Size-Based Queuing - An Approach to Improve Bandwidth 28 Utilization in TSN Networks

  29. Analysis - Latency • Messages are distributed into queue-set, message has to compete against less other messages inside a queue  Message reaches front of the queue faster 2  Latency decreases . 1 Size-Based Queuing - An Approach to Improve Bandwidth 29 Utilization in TSN Networks

  30. Analysis - Latency • (depending on arbitration scheme) • Arbitration within queue-sets affects latency of all LP messages 33  Latency of lower relative priority messages increases  Latency of higher relative priority messages decreases 2 1 Size-Based Queuing - An Approach to Improve Bandwidth 30 Utilization in TSN Networks

  31. Analysis - Latency • Actual number depend on SBQ parameters • Number of queues in a queue-set • Values of thresholds • Schedule/intra queue-set arbitration (algorithm, priorities) • Impact differs for each queue of the queue-set 2 3 33 33 2 2 1 1 3 1 Size-Based Queuing - An Approach to Improve Bandwidth 31 Utilization in TSN Networks

Recommend


More recommend