WirelessHART Modeling and Performance Evaluation

Transcription

1 WirelessHART Modeling and Performance Evaluation Anne Remke and Xian Wu October 24, 2013 A. Remke and X. Wu (University of Twente) WirelessHART October 24, / 21

2 WirelessHART [ A. Remke and X. Wu (University of Twente) WirelessHART October 24, / 21

3 Introduction Control applications switch from wired communication to Wireless reduced installation costs and increased flexibility can be adapted during runtime involves multiple communication hops Challenge is the predictability! Recent work by Alur, Pappas et al. introduces formal syntax and semantics for multi-hop WirelessHart different kinds of link failure models sufficient conditions for stability in control loop if only single link fails A. Remke and X. Wu (University of Twente) WirelessHART October 24, / 21

4 Introduction Control applications switch from wired communication to Wireless reduced installation costs and increased flexibility can be adapted during runtime involves multiple communication hops Challenge is the predictability! Recent work by Alur, Pappas et al. introduces formal syntax and semantics for multi-hop WirelessHart different kinds of link failure models sufficient conditions for stability in control loop if only single link fails A. Remke and X. Wu (University of Twente) WirelessHART October 24, / 21

5 This work Proposes to separate failure model from control analysis allows to consider the possible failure of all involved links Computes reachability probabilities for certain messages, paths or networks evaluates influence of different parameters takes into account different failure models could be included into control analysis directly A. Remke and X. Wu (University of Twente) WirelessHART October 24, / 21

6 This work Proposes to separate failure model from control analysis allows to consider the possible failure of all involved links Computes reachability probabilities for certain messages, paths or networks evaluates influence of different parameters takes into account different failure models could be included into control analysis directly A. Remke and X. Wu (University of Twente) WirelessHART October 24, / 21

7 WirelessHART protocol feedback control field devices are source and relay nodes gateway is routing destination pseudo-random frequency channel hopping and channel blacklisting to avoid channel overlaping and to reduce interference network layer determines routing upstream and downstream graph routing data link layer defines strict 10 ms time slots Time Division Multiple Access (TDMA) for collision-free and deterministic communications one transmission per slot MAC layer is slotted and synchronized series of consecutive slots forms super-frame A. Remke and X. Wu (University of Twente) WirelessHART October 24, / 21

8 WirelessHART protocol feedback control field devices are source and relay nodes gateway is routing destination pseudo-random frequency channel hopping and channel blacklisting to avoid channel overlaping and to reduce interference network layer determines routing upstream and downstream graph routing data link layer defines strict 10 ms time slots Time Division Multiple Access (TDMA) for collision-free and deterministic communications one transmission per slot MAC layer is slotted and synchronized series of consecutive slots forms super-frame A. Remke and X. Wu (University of Twente) WirelessHART October 24, / 21

9 WirelessHART protocol feedback control field devices are source and relay nodes gateway is routing destination pseudo-random frequency channel hopping and channel blacklisting to avoid channel overlaping and to reduce interference network layer determines routing upstream and downstream graph routing data link layer defines strict 10 ms time slots Time Division Multiple Access (TDMA) for collision-free and deterministic communications one transmission per slot MAC layer is slotted and synchronized series of consecutive slots forms super-frame A. Remke and X. Wu (University of Twente) WirelessHART October 24, / 21

10 WirelessHART protocol feedback control field devices are source and relay nodes gateway is routing destination pseudo-random frequency channel hopping and channel blacklisting to avoid channel overlaping and to reduce interference network layer determines routing upstream and downstream graph routing data link layer defines strict 10 ms time slots Time Division Multiple Access (TDMA) for collision-free and deterministic communications one transmission per slot MAC layer is slotted and synchronized series of consecutive slots forms super-frame A. Remke and X. Wu (University of Twente) WirelessHART October 24, / 21

11 Radio connectivity graph Wireless Control Netowork Plant s1 a1 s2 a2 v1 v2 v4 v5 controller s3 a3 v3 v6 v7 A. Remke and X. Wu (University of Twente) WirelessHART October 24, / 21

12 Model information flow Wireless Control Netowork Plant s1 a1 s2 a2 v1 v2 v4 v5 controller s3 a3 v3 v6 v7 Static routing v 1 v 4 controller, v 2 v 5 controller, v 3 v 6 v 7 controller Communication schedule with superframe of size 7 η = ( v 1, v 4, v 2, v 5, v 3, v 6, v 4, c, v 5, c, v 6, v 7, v 7, c ). A. Remke and X. Wu (University of Twente) WirelessHART October 24, / 21

13 Model information flow Wireless Control Netowork Plant s1 a1 s2 a2 v1 v2 v4 v5 controller s3 a3 v3 v6 v7 Static routing v 1 v 4 controller, v 2 v 5 controller, v 3 v 6 v 7 controller Communication schedule with superframe of size 7 η = ( v 1, v 4, v 2, v 5, v 3, v 6, v 4, c, v 5, c, v 6, v 7, v 7, c ). A. Remke and X. Wu (University of Twente) WirelessHART October 24, / 21

14 WirelessHART parameters Super-frame all field nodes share the same super-frame of size (F s ) slots are specifically allocated to field devices to transmit uplink/downlink Reporting interval frequency at which measurements are taken and communicated (I s ) reduced frequency saves wireless communiciation overhead and extends life-time of field devices Message life cycle Time-to-Live (TTL) defines message life time TTL is decreased with each slot (uplink OR downlink) A. Remke and X. Wu (University of Twente) WirelessHART October 24, / 21

15 WirelessHART parameters Super-frame all field nodes share the same super-frame of size (F s ) slots are specifically allocated to field devices to transmit uplink/downlink Reporting interval frequency at which measurements are taken and communicated (I s ) reduced frequency saves wireless communiciation overhead and extends life-time of field devices Message life cycle Time-to-Live (TTL) defines message life time TTL is decreased with each slot (uplink OR downlink) A. Remke and X. Wu (University of Twente) WirelessHART October 24, / 21

16 WirelessHART parameters Super-frame all field nodes share the same super-frame of size (F s ) slots are specifically allocated to field devices to transmit uplink/downlink Reporting interval frequency at which measurements are taken and communicated (I s ) reduced frequency saves wireless communiciation overhead and extends life-time of field devices Message life cycle Time-to-Live (TTL) defines message life time TTL is decreased with each slot (uplink OR downlink) A. Remke and X. Wu (University of Twente) WirelessHART October 24, / 21

17 Hierarchical path model Information flow each flow periodically generates packets at source passes it through to its destination Time-triggered nature of the protocol iterate through slots of superframe and schedule possible transitions idle slot (if communication schedule does not allow a transmission) successful or unsuccessful transmission Compositional model per path state reflects age of the message at each hop (age 1, age 2,..., age n ) unique initial state (1, 0, 0,..., 0) allows to include failure probabilities A. Remke and X. Wu (University of Twente) WirelessHART October 24, / 21

18 Hierarchical path model Information flow each flow periodically generates packets at source passes it through to its destination Time-triggered nature of the protocol iterate through slots of superframe and schedule possible transitions idle slot (if communication schedule does not allow a transmission) successful or unsuccessful transmission Compositional model per path state reflects age of the message at each hop (age 1, age 2,..., age n ) unique initial state (1, 0, 0,..., 0) allows to include failure probabilities A. Remke and X. Wu (University of Twente) WirelessHART October 24, / 21

19 Underlying link model for transient failures p fl 1-p fl UP DOWN 1-p rc p rc successful transmission of each bit with probability 1 BER for WirelessHART message with L bits results in failure proability of p fl = 1 (1 BER) L recovery p rc = 0.9 for transient failures For links in transient state: For links in steady-state: [p s (t), p f (t)] = p(t) = p(0) [ ] t 1 pfl p fl p rc 1 p rc p rc p fl [p s, p f ] = [π(up), π(down)] = [, ] p rc + p fl p rc + p fl A. Remke and X. Wu (University of Twente) WirelessHART October 24, / 21

20 Underlying link model for transient failures p fl 1-p fl UP DOWN 1-p rc p rc successful transmission of each bit with probability 1 BER for WirelessHART message with L bits results in failure proability of p fl = 1 (1 BER) L recovery p rc = 0.9 for transient failures For links in transient state: For links in steady-state: [p s (t), p f (t)] = p(t) = p(0) [ ] t 1 pfl p fl p rc 1 p rc p rc p fl [p s, p f ] = [π(up), π(down)] = [, ] p rc + p fl p rc + p fl A. Remke and X. Wu (University of Twente) WirelessHART October 24, / 21

21 Example DTMC for 3-hop path Reporting interval is I s = 1, uplink frame-size F up = 7 and communication schedule η = (, n 1, n 2,,, n 2, n 3,, n 3, G ) R7 p s3 1 6,6,6 7,7,7 p s ,3,- 4,4,- 5,5,- 6,6,- 1 7,7,- 1 1,-,- 2,-,- 3,-, ,-,- 5,-,- 6,-,- 1 7,-,- p f1 p s2 p f2 1 1 p f3 Discard A. Remke and X. Wu (University of Twente) WirelessHART October 24, / 21

22 Predictability measures Reachability probability that message generated at source reaches gateway before end of given reporting interval Delay time difference between T born and T rec, which equals the age of a message delay distribution τ can be derived from transient distribution of the DTMC Utilization indicates fraction of slots that transmitted irrespective of success directly relates to network communication overhead and power consumption A. Remke and X. Wu (University of Twente) WirelessHART October 24, / 21

23 Predictability measures Reachability probability that message generated at source reaches gateway before end of given reporting interval Delay time difference between T born and T rec, which equals the age of a message delay distribution τ can be derived from transient distribution of the DTMC Utilization indicates fraction of slots that transmitted irrespective of success directly relates to network communication overhead and power consumption A. Remke and X. Wu (University of Twente) WirelessHART October 24, / 21

24 Predictability measures Reachability probability that message generated at source reaches gateway before end of given reporting interval Delay time difference between T born and T rec, which equals the age of a message delay distribution τ can be derived from transient distribution of the DTMC Utilization indicates fraction of slots that transmitted irrespective of success directly relates to network communication overhead and power consumption A. Remke and X. Wu (University of Twente) WirelessHART October 24, / 21

25 Typcial WirelessHART network Real plant settings according to HART Communication Foundation 30% of nodes communicate directly with gateway access points about 50% are two hops away remaining 20% may be 3 or 4 hops away Link layer availability WirelessHART MAC layer payload length is 127 bytes for BER = it follows p fl = and π(up) = A. Remke and X. Wu (University of Twente) WirelessHART October 24, / 21

26 Typcial WirelessHART network Real plant settings according to HART Communication Foundation 30% of nodes communicate directly with gateway access points about 50% are two hops away remaining 20% may be 3 or 4 hops away Link layer availability WirelessHART MAC layer payload length is 127 bytes for BER = it follows p fl = and π(up) = A. Remke and X. Wu (University of Twente) WirelessHART October 24, / 21

27 Connectivity graph of a typcial WirelessHART network A. Remke and X. Wu (University of Twente) WirelessHART October 24, / 21

28 Connectivity graph of a typcial WirelessHART network Parameters reporting interval is 4 superframe size for uplink is 20 η a = ( n1, G, n2, G, n3, G, n4, n1, n1, G, n5, n1, n1, G, n6, n2, n2, G, n7, n3, n3, G, n8, n3, n3, G, n9, n6, n6, n2, n2, G, n10, n7, n7, n3, n3, G ) A. Remke and X. Wu (University of Twente) WirelessHART October 24, / 21

29 Robustness against transient link failures transient UP probability steady state probability when p =0.184 fl transient UP probability when p =0.184 fl steady state probability when p fl =0.05 transient UP probability when p fl = Time slot A. Remke and X. Wu (University of Twente) WirelessHART October 24, / 21

30 Reachability Table: The reachability probabilities with a link failure lasting one cycle Path number Hop number Reachability (%) without link failure with link failure Bottleneck The longest path with the lowest availability forms the bottleneck of the system. A. Remke and X. Wu (University of Twente) WirelessHART October 24, / 21

31 Reachability Table: The reachability probabilities with a link failure lasting one cycle Path number Hop number Reachability (%) without link failure with link failure Bottleneck The longest path with the lowest availability forms the bottleneck of the system. A. Remke and X. Wu (University of Twente) WirelessHART October 24, / 21

32 Utilization Table: Influence of π(up) on the utilization rate of the example network Link availability Utilization rate Bad links not only degrade the control stability but also introduce more communication overhead and power consumption. A. Remke and X. Wu (University of Twente) WirelessHART October 24, / 21

33 Utilization Table: Influence of π(up) on the utilization rate of the example network Link availability Utilization rate Bad links not only degrade the control stability but also introduce more communication overhead and power consumption. A. Remke and X. Wu (University of Twente) WirelessHART October 24, / 21

34 Overall delay distribution probability delay (ms) A. Remke and X. Wu (University of Twente) WirelessHART October 24, / 21

35 Expected delays with different schedules Schedule η a X= 10 Y= Schedule η b X= 7 Y= expected delays (ms) path A. Remke and X. Wu (University of Twente) WirelessHART October 24, / 21

36 Reachability probabilities with I s = 2 and I s = 4 A. Remke and X. Wu (University of Twente) WirelessHART October 24, / 21

37 Conclusions WirelessHart is able to deliver reliable service under typical industrial environments important to choose parameters appropriate to achieve stable control Contribution Tool to automatically derive the DTMC for a specific network, communication schedule, routing graph, and reporting interval and to compute predictability measures reachability delay distribution utilization A. Remke and X. Wu (University of Twente) WirelessHART October 24, / 21

38 Conclusions WirelessHart is able to deliver reliable service under typical industrial environments important to choose parameters appropriate to achieve stable control Contribution Tool to automatically derive the DTMC for a specific network, communication schedule, routing graph, and reporting interval and to compute predictability measures reachability delay distribution utilization A. Remke and X. Wu (University of Twente) WirelessHART October 24, / 21

39 Conclusions WirelessHart is able to deliver reliable service under typical industrial environments important to choose parameters appropriate to achieve stable control Contribution Tool to automatically derive the DTMC for a specific network, communication schedule, routing graph, and reporting interval and to compute predictability measures reachability delay distribution utilization A. Remke and X. Wu (University of Twente) WirelessHART October 24, / 21

40 Conclusions WirelessHart is able to deliver reliable service under typical industrial environments important to choose parameters appropriate to achieve stable control Contribution Tool to automatically derive the DTMC for a specific network, communication schedule, routing graph, and reporting interval and to compute predictability measures reachability delay distribution utilization A. Remke and X. Wu (University of Twente) WirelessHART October 24, / 21