Alternative Shaper for Scheduled Traffic in Time Sensitive Networks

Size: px

Start display at page:

Download "Alternative Shaper for Scheduled Traffic in Time Sensitive Networks"

Crystal Waters
6 years ago
Views:

Alternative Shaper for Scheduled Traffic in Time Sensitive Networks 213-1-15 IEEE 82.

1 Alternative Shaper for Scheduled Traffic in Time Sensitive Networks IEEE 82.1 TSN TG Meeting - Vancouver Franz-Josef Götz, Siemens AG franz-josef.goetz@siemens.com

2 Structure of this Presentation 1. Feature Diagram for Time Sensitive Networks Industrial 2. Traffic Shapers to support Low Latency for Scheduled Traffic (ST) 3. Simulation Results (one Use Case) Time aware Shaper (TAS) <-> Burst limiting Shaper (BLS) a) Low Latency for Scheduled Traffic with constant Frame Size b) Low Latency for Scheduled Traffic with random Frame Size c) Low Latency for Scheduled Traffic with random Frame Size and optimized Window Size for TAS Seite

3 Feature Diagram for Time Sensitive Networks Industrial - Consideration Topology complexity (line meshed) Number of nodes (1.. 2) Number of communication relations (1.. 4) Converged network (efficiently usage of available bandwidth) Seamless availability (two e2e path.. single point of failure per segment e.g. ring) Flexibility (low.. high) Robustness (low.. high) Configuration effort (low.. high) Seite Latency (2ms.. 8µs over 7 hops) Synchronization (accuracy <1µs,, two time scales)

4 Topologies, Redundancy, Synchronization & Communication Relations in Industrial Networks (1) BR IO-C bridge end station IO-Controller (PLC) BR GPS GM Univ. Time IO-C IO-Controller & Gand Master evice (Actuator or sensor as bridged end station) BR BR BR BR IO-C GM Func. Unit 1 BR IO-C GM Func. Unit 2 IO-C GM Func. Unit 3 IO-C GM Func. Unit 4 ring topologie BR IO-C GM IO-C GM Func. Unit 7 IO-C GM BR Func. Unit 5 IO-C GM by application coupled rings IO-C GM Func. Unit 6 comb topologie line topologie BR BR redundant ring topologie Seite

5 Future Topologies, Redundancy, Synchronization & Communication Relations in Industrial Networks (1) Seite

6 Feature Diagram for Time Sensitive Networks Industrial Topology complexity (line meshed) Number of nodes (1.. 2) Number of communication relations (1.. 4) Energy automation control Converged network (efficiently usage of available bandwidth) medium industrial control (cycle <=5ms) Seamless availability (two e2e path.. single point of failure per segment e.g. ring) medium speed motion control (cycle <=1ms) Flexibility (low.. high) Robustness (low.. high) Configuration effort (low.. high) Seite Latency (2ms.. 8µs over 7 hops) high speed motion control (cycle <=125µs) Synchronization (accuracy <1µs,, two time scales)

7 Feature List for Time Sensitive Networks Industrial Feature Technology Applications High speed motion control Medium speed motion control Medium Industrial control Energy Automation Automotive???... Sync Working Clock Hot- or cold- standby GM Multi path for sync messages to support high availability Sync tree for universal time Seamless Availability Scheduling Reservation L2 Routing Network segment protection Duplicate generation & elimination (end station & bridges) Time based transmission in nodes (offline engineered) Bandwidth & resources Multi path - without reconfiguration, at startup Single path with reconfiguation () () Low Latency Shaper Time Aware Shaper (TAS) Burst Limiting Shaper (BLS) Pre-emption () Scheduled Traffc Class A & B Seite

8 Structure of this Presentation 1. Feature Diagram for Time Sensitive Networks Industrial 2. Traffic Shapers to support Low Latency for Scheduled Traffic (ST) 3. Simulation Results (one Use Case) Time aware Shaper (TAS) <-> Burst limiting Shaper (BLS) a) Low Latency for Scheduled Traffic with constant Frame Size b) Low Latency for Scheduled Traffic with random Frame Size c) Low Latency for Scheduled Traffic with random Frame Size and optimized Window Size for TAS Seite

9 Goals for successful Time Sensitive Networks Adapt complexity to requirements (KISS) Minimize overall complexity at application side to improve acceptance Minimization of hardware complexity should not lead to unnecessary higher complexity for network configuration management and end users Seite

10 Recap: Low Latency is required to minimize Transmission Time for Scheduled Traffic Device Controller - Communication Controller Device -Communication Transmission of Scheduled Traffic within transmission time for Scheduled Traffic Seite

11 Recap: Proposed Traffic Shapers to support Low Latency for Scheduled Traffic (ST) Details: Time aware Shaper (TAS) MU MU Time MU Pre-emption NEW: Burst Limiting Shaper (BLS) MU MU MU Pre-emption Seite

12 Recap: Why an ADDITIONAL Shaper for Scheduled Traffic Reasons: Shaper with less configuration effort in bridges (only bandwidth not window size) Adapt to different link speeds within a network Automatically adaption to traffic flow For mashed networks with multiple talker Plugging of new components may change bandwidth but the time schedule must not adapted The shaper shall support periodic and event-based Scheduled Traffic (bandwidth limitation) Robust for safety critical applications Emergency operation features (fail operational behavior) Seite

13 Requirements on a Traffic Shaper for Scheduled Traffic Guaranteed Low Latency for Scheduled Traffic Guarantee for maximum burst size for Scheduled Traffic Guaranteed bandwidth for other traffic classes Avoid to add additional latencies Seite

14 Proposed: Burst Limiting Shaper (BLS) for Scheduled Traffic number of queued control data frames ULL_2 ULL_1 ULL_2 ULL_3 ULL_3 ULL-Queue Priority Legacy-Queue Priority ULL_4 ULL_6 ULL_5 ULL_6 ULL-Queue Priority < Legacy-Queue Priority ULL_7 t number of queued legacy frames transmitted frames Leg_1 Leg_2 Leg_3 Leg_4 t ULL_1 (496 Bit) ULL_2 (372 Bit) ULL_3 (248 Bit) Leg_1 (372 Bit) ULL_4 (8192 Bit) ULL_5 (7168 Bit) ULL_6 (248 Bit) Leg_2 (248 Bit) Leg_3 (248 Bit) ULL_7 (248 Bit) t Bits MaxLevel ULL_1 ResumeLevel SlopeSend ULL_2 ULL_3 SlopeIdle ULL_4 ULL_5 SlolpenSend SlopeSend ULL_6 SlolpIdle Resume_Level ULL_7 t Seite

15 Recap: Pre-emption to support Low Latency for Scheduled Traffic Mechanism to support pre-emption: LLDP shall be use for auto negotiation Port Pre-emption supported Min fragment size Configurable pre-emtive traffic classes Peer-to-Peer fragmentation of legacy traffic on egress port Peer-to-Peer reassemble of legacy traffic on ingress port legacy frame 1458 Byte 4 Bridge 15 Byte ULL Stream DA SA VLAN ET Preemption 64 Bridge legacy frame fragment 1522 Byte 15 Byte 5 ULL stream Port 82.1Q + Extentions for Reserved and Scheduled Traffic ST High ST Low AV SR A AV SR B Prio 7 Prio 4,1, B/T B/T P C C preemptive traffic classes preemptable traffic classes MAC B/T: BLS or TAS shaper for Scheduled Traffic C: Credit based Shaper for Reserved Traffic P: Priority based Shaper for legacy traffic Legend: 128 Bytes Peer DA ET FragFCS Payload FCS legacy Frame Reserved Multicast Address Ethertype of fragment FCS is signaling a fragment Fragment- Frames DA SA VLAN ET Payload Fragment 1 FragFCS Peer DA SA ET Payload Fragment 2 FragFCS Peer DA SA ET Payload Fragment 3 FCS FragLength (8Byte granular) Min. 64 Bytes Seite

16 Structure of this Presentation 1. Feature Diagram for Time Sensitive Networks Industrial 2. Traffic Shapers to support Low Latency for Scheduled Traffic (ST) 3. Simulation Results (one Use Case) Time aware Shaper (TAS) <-> Burst limiting Shaper (BLS) a) Low Latency for Scheduled Traffic with constant Frame Size b) Low Latency for Scheduled Traffic with random Frame Size c) Low Latency for Scheduled Traffic with random Frame Size and optimized Window Size for TAS Seite

17 Simulation - Latency for Scheduled Traffic Time awar Shaper (TAS)<-> Burst limiting Shaper (BLS) Topology & CR: Controller (PLC, GE) <-> Devices (Actuators, Sensors, FE) Seite

18 Simulations of 3 UC - Latency for Scheduled Traffic Time aware Shaper (TAS)<-> Burst limiting Shaper (BLS) General Settings (1): Network: 8 bridges, 8x8 devices (bridged end stations) Real time application (synchronized) Transmission order (): farthest first, nearest last Traffic load for Scheduled Traffic < Frame Size for Scheudeld Traffic: UC 1 constant size: 64 Bytes UC 2 random size: 1% 64 Bytes, 1% 512 Bytes, 8% between 128~384 Bytes Best effort traffic: Traffic load < 3% Frame size: 2 max_size, 2 min_size, between 25~125 Bytes 2 burst (frames in chain), 7 non-burst Seite

19 Simulations of 3 UC - Latency for Scheduled Traffic Time aware Shaper (TAS)<-> Burst Limiting Shaper (BLS) General Settings (2) BLS & TAS Transmission period: 25 us for Scheduled Traffic with constant frame size, Transmission period: 875 us for Scheduled Traffic with random frame size Window size (only for TAS): UC 1 Window size = 85 us for Scheduled Traffic with constant frame size UC 2 Window size is 4 us for Scheduled Traffic with random frame size UC 3 based on UC 2 Window size optimized Window start time always at the beginning of cycle Window close time varies for different location of station Window close time right after the station has transmitted the last Scheduled frame Cut-through mod only for Scheduled Traffic Delay: 48 bytes Bridging delay: 5 ns; cable + PHY delay: 75 ns Pre-emption in combination with TAS or BLS Seite

20 UC 1a Simulation - Low Latency for Scheduled Traffic with constant Frame Size of 64Bytes, S&F@GE & FE Shaper S&F (Store & GE & FE, Scheduled Traffic with constant frame size = 64 Byte Load Scheduled Traffic communication time * (us) (79.64) (79.64) GE E2E delay** (us) FE E2E delay** (us) (95.36) (123.4) 5.84 (57.55) 5.14 (64.86) (581.2) (652.9) 1% 84.9 (97.62) (123.5) (6.68) (69.66) (584.4) (624.3) BLS (1.4) (123.6) 6.46 (73.75) (79.56) (785.7) (723.2) 2% (1.8) (123.9) (91.1) 9.28 (94.27) (687.7) 47.5 (875.4) (11.4) (124.1) 1.24 (97.19) (148.3) (745.5) (994.) 3% (13.7) (124.4) (155.7) (161.7) (879.3) 6.99 (929.7) 2.56 (112.2) (19.2) (82.5) (714.1) TAS (cycle = 25 us, window size = 85 us) 1% 1 2% (79.64) (79.64) (122.) (196.4) (24.7) 2.68 (12.2) (172.2) 27.7 (221.3) (744.9) (929.4) (939.2) (749.2) (834.6) (943.) (265.4) (37.4) (968.) (1162) 3% (373.5) (379.1) 11.4 (999.9) (133) Value Format: mean (max) * transmission delay for Scheduled Traffic: time from start of transmission of the first frame to receiving of the last frame ** E2E delay for Traffic: time from transmitting the frame to receiving Seite

21 UC 1b Simulation - Low Latency for Scheduled Traffic with constant Frame Size of 64Bytes, S&F@GE + CT@FE Shaper Load S&F@GE, CT@FE, Scheduled Traffic with constant frame size = 64 Byte Scheduled Traffic communication time * (us) 7.89 (7.89) 7.89 (7.89) GE E2E delay** (us) FE E2E delay** (us) (88.39) (92.72) 5.84 (57.55) (66.11) 4.53 (581.2) (62.1) 1% (91.5) (93.58) (6.58) (67.89) (584.4) 4.85 (63.9) BLS (91.73) (94.22) 6.46 (73.75) (77.92) (687.7) (71.6) 2% 81.9 (95.95) 8.82 (97.61) (91.1) (95.35) (745.5) 47.8 (856.4) (97.6) (99.5) 1.24 (97.19) (155.5) (785.7) 52.8 (898.) 3% (99.1) (13.1) (155.7) (159.4) (879.3) (962.5) 2.56 (112.2) (19.2) (82.5) (714.1) TAS (cycle = 25 us, window size = 85 us) 1% 1 2% 7.89 (7.89) 7.89 (7.89) (122.) (196.4) (24.7) 2.68 (12.2) (172.2) 27.7 (221.3) (744.9) (929.4) (939.2) (749.2) (834.6) (943.) (265.4) (37.4) (968.) (1162) 3% (373.5) (379.1) 11.4 (999.9) (133) Value Format: mean (max) Seite

22 UC 1c Simulation - Low Latency for Scheduled Traffic with constant Frame Size of 64Bytes, CT@GE & FE Scheduled Traffic with constant frame size = 64 Byte Shaper Load Scheduled Traffic communication time * (us) GE E2E delay** (us) FE E2E delay** (us) (69.99) (69.99) (88.13) (91.68) 5.8 (57.25) (66.1) 4.53 (581.2) (62.1) 1% (91.3) (91.89) (6.19) (67.74) (584.4) 4.85 (63.9) BLS (91.55) (93.54) (73.77) 7.65 (78.23) (687.7) (71.6) 2% 8.37 (95.95) 8.1 (97.3) (9.1) (95.4) (745.5) 47.8 (856.4) (97.1) (97.74) 1.24 (96.54) (154.2) (785.7) 52.8 (898.) 3% 82.2 (97.48) (12.2) (154.4) (158.6) (879.2) (962.5) 2.56 (112.2) (19.2) (82.5) (714.1) TAS (cycle = 25 us, window size = 85 us) 1% 1 2% (69.99) (69.99) (122.) (196.4) (24.7) 2.68 (12.2) (172.2) 27.7 (221.3) (744.9) (929.4) (939.2) (749.2) (834.6) (943.) (265.4) (37.4) (968.) (1162) 3% (373.5) (379.1) 11.4 (999.9) (133) Value Format: mean (max) Seite

23 Simulation - Low Latency for Scheduled Traffic with random Frame Size Distribution of frame size for Scheduled Traffic: Seite

24 UC 2a Simulation - Low Latency for Scheduled Traffic with random Frame Size S&F@GE & FE S&F@GE & FE, Scheduled Traffic with random frame size Shaper Load Scheduled Traffic communication time * (us) GE E2E delay** (us) FE E2E delay** (us) (367.2) (355.9) (38.2) 357. (365.4) (156.2) 4.57 (84.78) 53.5 (65.9) (638.4) 1% 369. (38.6) (369.8) (158.5) 5.3 (119.3) 5.89 (687.5) (676.7) BLS (38.9) (372.4) (161.) (153.93) (774.8) (749.5) 2% 37.2 (382.2) (373.4) (185.4) 8.5 (171.8) (778.) (792.2) (384.3) 36. (374.4) 11.6 (188.8) 1.82 (186.6) 63.2 (782.7) 61.6 (94.7) 3% (384.5) 36.5 (381.5) (196.3) 13.9 (23.1) (81.5) (995.6) 92.4 (43.2) (46.5) (922.7) (1127) TAS (cycle = 875 us, window size = 38 us) 1% 1 2% (367.2) (355.9) (47.8) 17.8 (418.7) (45.1) (46.6) 16.2 (418.8) (424.2) (193) (121) (1333) 16.7 (1162) 17.5 (1165) (1215) (485.2) (466.4) (1436) 27.5 (148) 3% 139. (51.3) (624.2) (1467) (1477) Value Format: mean (max) * transmission delay for Scheduled Traffic: time from start of transmission of the first frame to receiving of the last frame ** E2E delay for Traffic: time from transmitting the frame to receiving Seite

25 UC 2b Simulation - Low Latency for Scheduled Traffic with random Frame Size S&F@GE + CT@FE S&F@GE + CT@FE, Scheduled Traffic with random frame size Shaper NRT Load Scheduled Traffic communication time * (us) GE E2E delay** (us) FE E2E delay** (us) (248.8) (237.2) (255.4) (243.8) (156.2) (137.9) 5.52 (65.9) (67.1) 1% (255.5) (243.9) (158.5) (174.9) (687.5) 55.7 (672.9) BLS (255.6) (244.3) (161.) (177.4) 53.5 (778.) (68.7) 2% (255.8) (244.5) 9.6 (185.4) (18.5) (782.7) (785.7) (255.9) (244.8) (188.8) (196.9) (793.4) (898.4) 3% (256.) (245.) (196.3) (214.7) (837.7) (964.1) 92.4 (43.2) (46.5) (922.7) (1127) TAS (cycle = 875 us, window size = 38 us) 1% 1 2% (248.8) (237.2) (47.8) 17.8 (418.7) (45.1) (46.6) 16.2 (418.8) (424.2) (193) (121) (1333) 16.7 (1162) 17.5 (1165) (1215) (485.2) (466.4) (1436) 27.5 (148) 3% 139. (51.3) (624.2) (1467) (1477) Value Format: mean (max) Seite

26 UC 2c Simulation - Low Latency for Scheduled Traffic with random Frame Size CT@GE & FE CT@GE & FE, Scheduled Traffic with random frame size Shaper Load Scheduled Traffic communication time * (us) GE E2E delay** (us) FE E2E delay** (us) (243.7) (231.8) 244. (25.6) 232. (238.1) (154.9) 4.92 (144.5) 5.74 (65.9) (67.1) 1% (25.8) 232. (238.3) (156.9) (173.8) (687.5) 55.7 (672.9) BLS (25.8) (238.4) ) (175.4) (778.) (68.7) 2% (25.9) (238.5) (184.5) (18.6) (782.7) (785.7) (25.9) (238.5) (186.9) 12.4 (198.1) (793.4) (898.4) 3% (251.) (238.6) (194.4) (216.2) (937.7) (964.1) 92.4 (43.2) (46.5) (922.7) (1127) TAS (cycle = 875 us, window size = 38 us) 1% 1 2% (243.7) (231.8) (47.8) 17.8 (418.7) (45.1) (46.6) 16.2 (418.8) (424.2) (193) (121) (1333) 16.7 (1162) 17.5 (1165) (1215) (485.2) (466.4) (1436) 27.5 (148) 3% 139. (51.3) (624.2) (1467) (1477) Value Format: mean (max) Seite

27 UC 3 Simulation - Low Latency for Scheduled Traffic with random Frame Size CT@GE & FE and optimized Window Size (close) Shaper Load CT@GE & FE, Scheduled Traffic with random frame Size + optimized window size Scheduled Traffic communication time * (us) GE E2E delay** (us) FE E2E delay** (us) TAS Not optimized window size (cycle = 875 us, window size = 38 us) 1% 1 2% (43.2) (47.8) 17.8 (418.7) (45.1) (485.2) (46.5) (46.6) 16.2 (418.8) (424.2) (466.4) (922.7) (193) (121) (1333) (1436) (1127) 16.7 (1162) 17.5 (1165) (1215) 27.5 (148) 3% (243.7) (231.8) 139. (51.3) (624.2) (1467) (1477) TAS Optimized window size (cycle = 875 us, window size < 38 us) 1% 1 2% (267.1) 34.6 (269.2) (277.1) (292.6) 45.6 (296.8) (254.8) (255.4) 4.37 (267.5) (275.2) 49.7 (31.7) (65.5) (776.3) (793.7) (9.6) (817.) (591.6) 59.3 (6.2) 6.2 (682.2) (79.3) (817.6) 3% 5.94 (322.7) (352.1) (92.1) (971.9) Value Format: mean (max) Seite

28 Comparison BLS <-> TAS for Scheduled Traffic with random Frame Size & FE and optimized Window Size Shaper BLS <-> TAS with optimized window size and Schedule Traffic with random frame size Load Scheduled Traffic communication time * (us) (243.7) (231.8) GE E2E delay** (us) FE E2E delay** (us) 244. (25.6) 232. (238.1) (154.9) 4.92 (144.5) 5.74 (65.9) (67.1) 1% (25.8) 232. (238.3) (156.9) (173.8) (687.5) 55.7 (672.9) BLS (25.8) (238.4) ) (175.4) (778.) (68.7) 2% (25.9) (238.5) (184.5) (18.6) (782.7) (785.7) (25.9) (238.5) (186.9) 12.4 (198.1) (793.4) (898.4) 3% (251.) (238.6) (194.4) (216.2) (937.7) (964.1) TAS Optimized window size (cycle = 875 us, window size < 38 us) 1% 1 2% (243.7) (231.8) 31.1 (267.1) 34.6 (269.2) (277.1) (292.6) 45.6 (296.8) (254.8) (255.4) 4.37 (267.5) (275.2) 49.7 (31.7) (65.5) (776.3) (793.7) (9.6) (817.) (591.6) 59.3 (6.2) 6.2 (682.2) (79.3) (817.6) 3% 5.94 (322.7) (352.1) (92.1) (971.9) Value Format: mean (max) Seite

29 Conclusion We have to find out which mechanism must be supported by TSN to cover the requirements for a wide range of industrial applications so that time sensitive Ethernet based networks (TSN) become successful!! Estimated change of Industrial Market Estimated ports Source: Contributions from Hirschmann, Siemens and Broadcom Seite

Robustness for Control-Data-Traffic in Time Sensitive Networks

Robustness for Control-Data-Traffic in Time Sensitive Networks 2013-07-15 -v01- IEEE 802.1 TSN TG Meeting Geneva - Switzerland Presenter: Franz-Josef Goetz, Siemens AG franz-josef.goetz@siemens.com Structure