Real-Time Wireless Routing for Industrial Internet of Things

Transcription

1 Real-Time Wireless Routing for Industrial Internet of Tings Cengjie Wu, Dolvara Gunatilaka, Mo Sa, Cenyang Lu Cyber-Pysical Systems Laboratory, Wasington University in St. Louis Department of Computer Science, State University of New York at Bingamton Abstract Wit te emergence of te Industrial Internet of Tings (IIoT), process industries ave started to adopt wireless sensor-actuator networks (WSANs) for control applications. It is crucial to acieve real-time communication in tis emerging class of networks, and routing as significant impacts on end-to-end communication delays in WSANs. However, despite considerable researc on real-time transmission sceduling and delay analysis for suc networks, real-time routing remains an open question for WSANs. Tis paper presents a conflict-aware real-time routing approac for WSANs, leveraging a key observation tat conflicts among transmissions involving a common field device can contribute significantly to communication delays in industrial WSANs, suc as WirelessHART networks. By incorporating conflict delays into te routing decisions, conflict-aware real-time routing algoritms allow a WSAN to accommodate more realtime flows wile meeting teir deadlines. Bot evaluations based on simulations and experiments on a pysical WSAN testbed sow tat conflict-aware real-time routing can lead to as muc as a tree-fold improvement in te real-time capacity of WSANs. I. INTRODUCTION Industrial networks connect sensors and actuators in te industrial facilities, suc as steel mills, oil refineries, and cemical plants, implementing complex monitoring and control processes. Te industrial Internet of Tings (IIoT) is a key enabling tecnology to realize te vision of Industry 4.. Wireless sensor-actuator networks (WSANs) provide an appealing solution to connect IIoT devices because tey require minimal infrastructure. Moreover, wireless modules can be used to easily and inexpensively retrofit existing sensors and actuators in industrial facilities, witout running cabling for communication and power [1]. Recent years ave witnessed world-wide deployment of WSANs implementing industrial standards, suc as WirelessHART [2] and ISA1.11a [3]. Feedback control loops for industrial automation impose stringent end-to-end delay requirements on data communication. To support a feedback control loop, te network periodically delivers data from sensors to a controller and ten delivers control commands to te actuators witin an end-toend deadline. Te effects of deadline misses in data communication may range from production inefficiency, equipment destruction, system failure for instability, to irreparable financial and environmental damage. Real-time communication in industrial WSANs is callenging due to teir limited bandwidt and multi-op mes topologies [1]. Furtermore, industrial standards, suc as WirelessHART and 6TiSCH [4] employ Time Slotted Cannel Hopping (TSCH), a TDMA-based MAC wit cannel opping, to acieve predictable and reliable communication. Endto-end communication delays in suc networks are affected by conflicts between transmissions involving a common device [5]. Because conflicting transmissions cannot be sceduled in a same time slot, transmission conflicts contribute significantly to te end-to-end communication delays of data flows. Recent studies demonstrated tat end-to-end delays of multi-op flows are eavily influenced by teir routes [6]. Wile real-time routing as received attention in te researc community, tere as been limited work on routing algoritms specifically designed for recent industrial WSAN standards. Moreover, tey generally ignore transmission conflicts in routing decisions, wic negatively affect te ability to meet te delay requirements of a large number of real-time flows. To meet tis open callenge, we propose conflict-aware routing, a novel approac to real-time routing in WirelessHART networks, a WSAN standard widely adopted in process industries. Te key novelty of conflict-aware routing is tat it incorporates transmission conflicts and sceduling into its routing decisions to improve real-time performance. Experiments on a pysical testbed and in numerical simulations sow tat conflict-aware routing can lead to as muc as a tree-fold improvement in te real-time capacity of a WSAN. Te rest of te paper is organized as follows. Section II reviews related work. Section III presents te network model. Section IV discusses te problem formulation. Section V provides a brief review of te existing delay analyses. Section VI presents our real-time routing algoritms. Section VII evaluates our routing algoritms troug experiments and simulations. Section VIII concludes te paper. II. RELATED WORK Te field of wireless sensor networks produced a multitude of sopisticated routing protocols (e.g., RPL [7] and ORPL [8], just to name a few). Designed for general-purpose applications tat do not demand real-time performance, tese routing protocols were optimized for efficiency and adaptivity to link dynamics. Tere were efforts to improve te real-time performance of traditional sensor networks in a best-effort manner suc as te works presented in [9] [12]. A common approac adopted by tese protocols is to employ localized algoritms to dynamically select te next op to forward a packet. However, tose decentralized approaces cannot

2 provide end-to-end delay guarantees and are incompatible wit recent standards for industrial WSANs. In contrast to traditional sensor networks, industrial WSAN standards adopt drastically different design coices in order to meet te stringent reliability and real-time requirements of IIoT. For example, in a WirelessHART network, links used for routing are usually more reliable and stable tan tose in traditional sensor networks. Furtermore, to acieve predictable latency, WirelessHART employs TSCH MAC protocol. Finally, WirelessHART adopts a centralized network manager responsible for computing routes for all flows in te network. Industrial WSANs tus need a new class of routing algoritms. Several groups proposed algoritms [13] [15] for reliable grap routing, a multi-pat routing approac supported by WirelessHART. Many efforts [16], [17] were geared toward improving te reliability and robustness of industrial WSANs. Oter routing algoritms suc as [18], [19] aimed to improve energy efficiency and prolong network lifetime. However, tese aforementioned algoritms were not targeted at improving te real-time performance of industrial WSANs. Te recent 6TiSCH standard combines te RPL routing protocol and te TSCH MAC to support decentralized adaptation [2]. Tis work is in contrast to our work tat is based on TSCH wit centralized sceduling. Tere exist real-time routing protocols for TDMA-based wireless sensor networks. Xu et al. [21] designed te PRTR protocol to minimize te delay of real-time traffic in a TDMAbased network. However, teir end-to-end delay bounds are probabilistic, wic is in contrast to many industrial applications tat require deterministic delay bounds. Nirjon et al. [22] proposed IAA, a real-time routing algoritm tat can guarantee end-to-end delay in TDMA-based networks. IAA employs euristics to assign sorter pats to flows wit tigter deadlines. It does not take into account transmission conflicts, wic play a significant role in communication delays in WirelessHART networks. Moreover, in contrast to our work tat is based on TSCH and WirelessHART, IAA is designed for a single-cannel TDMA network tat allows concurrent transmissions on te same cannel. III. NETWORK MODEL We consider a network model based on te WirelessHART standard [2] tat as been widely adopted in process industries. A WSAN consists of a gateway, multiple access points, and a set of field devices (e.g., sensors or actuators). Te access points and network devices are equipped wit alf-duplex radio transceivers compatible wit te IEEE pysical layer; togeter tey form a wireless mes network. A WSAN can use up to 16 cannels, as specified in te IEEE standard. Te access points are wired to te gateway and serve as bridges between te gateway and field devices. Te WSAN adopts a centralized network management approac, were te network manager (i.e., a software module running on te gateway or a ost connected to te gateway) manages all devices. Te network manager gaters te network topology information, and ten generates and disseminates te routes and transmission scedule to all network devices. Tis centralized network management arcitecture, adopted by te WirelessHART standard, enances te predictability and visibility of network operations at te cost of scalability. Te WSAN adopts te TSCH MAC on top of te IEEE pysical layer. TSCH is a TDMA-based protocol in wic all devices in te network are time syncronized. Time is divided into 1 ms slots, and eac slot can accommodate one packet transmission and its acknowledgment. In a slot, only one transmission is sceduled on eac cannel across te entire network to avoid cannel contention, and enance reliability. Moreover, TSCH supports cannel opping (i.e., eac node switces to a new receiving cannel in every time slot) to enance network resiliency troug cannel diversity. Te network operator can blacklist cannels wit poor quality. Te WirelessHART standard supports two types of routing: source routing and grap routing. Source routing provides a single route from a source to a destination, wereas grap routing provides multiple redundant routes in a routing grap. Hence, grap routing promotes reliability troug route diversity, at te cost of longer latency and iger energy cost [6]. Given our interest in real-time communication, tis paper focuses on source routing. In addition, wile our algoritms are designed for te WirelessHART standard, te insigts and approac may be extended to oter WSANs based on TSCH. IV. PROBLEM FORMULATION We consider a WSAN wit a set of N real-time flows F = {F 1,F 2,,F N }. For eac flow F k =(s k,d k, k,d k,t k ),a source s k generates a packet at a constant period T k. A packet must be delivered to a destination d k troug a source route k witin a relative deadline D k. Due to its simplicity and efficiency, fixed priority sceduling is commonly adopted as te real-time sceduling policy in CPU and traditional real-time networks (e.g., Control- Area Networks). A recent study [23] as sown tat fixed priority sceduling is an effective policy for real-time flows in WSANs. Hence, we will adopt te fixed priority sceduling framework in tis work. In practice, priorities are assigned based on deadlines, periods, or te criticality of te real-time flows. Priorities of flows remain constant during run-time unless te user requirements or te traffic demands are canged. In tis work, we use te deadline-monotonic priority assignment policy, were flows wit closer deadlines are assigned wit iger priorities. Priorities of flows wit te same deadline are randomly assigned. Our routing algoritms can be applied to any fixed priority assignment. Under a fixed priority sceduling policy, te transmissions of te flows are sceduled in te following way. We assume tat all flows are ordered by priorities. Flow F i as a iger priority tan flow F j if and only if i<j. Starting from te igest priority flow, F 1, te following procedure is repeated for every flow F i in decreasing order of priority. Te network manager scedules transmissions of te current flow F i in

3 te earliest available time slots and on available cannels. A time slot is available if no conflicting transmission is already sceduled in tat slot. Te goal of our routing algoritm is to find routes for te flows so tat every flow can meet its deadline. Te sortest pat algoritms based on op count [13] [15] are commonly adopted in WSANs. However, as sown in our simulation results, te effectiveness of tese algoritms is far from optimal. Based on te insigts from end-to-end delay analysis, we propose two new euristics to assign routes to meet real-time requirements. V. CONFLICT DELAY ANALYSIS In tis section, we summarize te delay analysis for WSANs. We later use tese insigts to design our routing algoritms. According to te previous work [23] on delay analysis, a packet can be delayed for two reasons: conflict delay and contention delay. Due to te alf-duplex radio, two transmissions conflict wit eac oter if tey sare a node (sender or receiver). In tis case, only one of tem can be sceduled in te current time slot. Terefore, if a packet conflicts wit anoter packet tat as already been sceduled in te current time slot, it as to be postponed to a later slot, resulting in conflict delay. Because a WSAN does not allow concurrent transmissions on te same cannel, eac cannel can accommodate only one transmission across te network in eac slot. If all cannels are assigned to transmissions of oter packets, a packet must be delayed to a later slot, resulting in contention delay. From te delay analyses presented in [23], [24], and our simulations, conflict delay plays a significant role in te endto-end delays of flows. Furtermore, routing directly impacts conflict delays, wereas contention delays largely depend on te number of cannels available. Terefore, in our routing design, we focus only on conflict delay. Saifulla et al. proposed Efficient Delay Analysis (EDA) [23], a state-of-teart delay analysis algoritm for WSANs. Here, we briefly discuss te EDA algoritm. We denote te total number of transmissions of flow F tat conflict wit flow F l as l. Here, flow F as a iger priority tan flow F l. l is counted based on te routes of te two flows. l equals te number of links in F s route tat sare nodes wit F l s route, times te number of transmissions sceduled on eac link. For example, given F s route is u! p! q! x! y, F l s route is v! p! q! z, and te number of transmissions over eac link is one, ten l = 3, i.e., tree links, {(u, p), (p, q), (q, x)}, in F s route sare nodes wit F l s route. Given a time interval of t slots, te number of packets of flow F tat contribute to te delay of a packet of flow F l during tis time interval is upper bounded by d t e, were is te period of flow, F. Terefore, te worst-case conflict delay of flow F l from all flows wit iger priority tan F l in a time interval t can be bounded as l (t) = X <l d t e l (1) Based on Equation (1), EDA uses an iterative fixed-point algoritm to get te upper bound of F l s conflict delay. We furter break down Equation (1) to learn ow muc a transmission of ig priority flow can delay a low priority flow. Here, we will give an approximation of conflict delay by a single transmission of ig priority flow. A packet of flow F l can be delayed only witin its lifetime D l (te relative deadline of flow F l ). To simplify Equation (1), we use F l s deadline as te lengt of te time window. We furter ignore te ceiling function and approximate te conflict delay tat F l can suffer from flow F as l = D l l (2) were l is te total conflict delay tat flow F brings to flow F l. Since te total number of transmissions of flow F tat conflict wit flow F l is l, we approximate te number of conflict delays from a single transmission of flow F as D l. We will use tis approximation in our routing design. VI. REAL-TIME ROUTING In tis section, we propose two real-time routing algoritms: Conflict-Aware Routing (CAR) and Iterative Conflict-Aware Routing (ICAR). A. Conflict-Aware Routing (CAR) As Section V sows, te conflict delay tat a single transmission of a ig priority flow, F, brings to a low priority flow, F l, is D l. We will incorporate tis idea into our Conflict- Aware Routing (CAR) algoritm, wic picks routes wit small conflict delays caused by ig-priority flows. Algoritm 1 presents te pseudocode of our CAR algoritm. Te two inputs to te algoritm are (1) a grap G(V,E), were V is te set of devices in te network and E is te set of links in te network, and (2) a flow set F = {F 1,F 2,,F N } ordered by priority. We assign routes for flows following te priority order, from te igest to te lowest. Eac link (u, v) as a link weigt w (u,v) and a delay coefficient c (u,v). For eac flow F l, we update te link weigts based on te routes of iger priority flows. If a link (u, v) sares at least one node wit a iger priority flow F s route, its weigt will be increased by D l, based on Equation (2). In our algoritm, we implement te link weigt update in two steps. In te first step, once te route R of a ig priority flow F is fixed, because every link on R will impose D l conflict delays on flows wit lower priority, we increase te delay coefficient of any link (u, v) tat sares at least one node wit R by 1. In te second step, wen we calculate te route for a lower priority flow F l, we update te weigt of eac link as w (u,v) = 1 + D l c (u,v), wic takes into account all flows tat ave iger priority. After updating te link weigts, we run Dijkstra s algoritm to find te pat l

4 wit te smallest pat weigt. Te algoritm terminates wen te flow wit te lowest priority is assigned a route N. Algoritm 1: Conflict-Aware Routing 1 Function CAR(G, F) Input : A grap G(V,E), A flow set F = {F 1,F 2,,F N } ordered by priority wit F l =(s l,d l,t l,d l ) Variable: link weigt w, link delay coefficient c Output : A route l for eac flow F l 2 for eac link (u, v) 2 E do 3 w (u,v) =1; 4 c (u,v) =; 5 for eac flow F l from F 1 to F n do 6 if l>1 ten 7 for eac link (u, v) 2 E do 8 w (u,v) =1+D l c (u,v) ; 9 Find te sortest pat l connecting s l to d l ; 1 Assign l as flow F l s route; 11 for eac link (u, v) 2 E do 12 if (u, v) sares at least one node wit F l s route l ten 13 c (u,v) = c (u,v) + 1 T l ; Now, we discuss te complexity of te CAR algoritm. We first ceck te complexity for eac flow (one iteration witin te for loop at lines 5-13). Te complexity to update te link weigts is O( E ). It takes O( E + V log V ) to execute te Dijkstra s algoritm, and O( E ) to update te delay coefficients. Ten, te total complexity of eac flow is O( E + V log V ). Finally, te complexity of our CAR algoritm is O(N( E + V log V )), were N is te number of flows in te WSAN. B. Iterative Conflict-Aware Routing (ICAR) By reducing te conflict delay of low priority flows, we can accommodate more flows wile meeting teir deadlines. However, CAR is based on flow priorities, and ig priority flows are not aware of te routes of low priority flows. We furter improve te real-time capacity by introducing an approac were ig priority flows also take into account te routes of low priority flows to avoid te overlapping routes. Tis approac gives low priority flows a iger cance to find routes tat are scedulable. Hence, in tis section, we introduce our Iterative Conflict-Aware Routing (ICAR) algoritm, wic is an extension of CAR ICAR is an iterative algoritm tat runs in rounds. Witin eac round, flows compute teir routes one by one. As wit CAR, eac flow F l will first update link weigts based on te routes of oter flows, and ten use Dijkstra s algoritm to find te pat wit te smallest pat weigt. ICAR ten determines if flow F l wit tis new route is scedulable under EDA. If yes, tis new route is assigned to F l, and flow F l is indicated as scedulable. Te delay coefficients of te links belonging to te old route of F l are deducted by 1 T l, and te coefficients of te links on te new route are increased by 1 T l. Oterwise, flow F l will not update its route. ICAR terminates wen (1) none of te flows update teir routes or, (2) all flows are scedulable under EDA or, (3) te number of iterations exceeds te preset tresold M. Oterwise, te algoritm will enter a new round. Te complexity of one round is O(N( E + V log V + D max )). Compared wit te complexity of te CAR algoritm O(N( E + V log V )), ICAR incurs an additional complexity of O(ND max ) for te delay analysis EDA, were D max is te maximum deadline in F. Te total complexity of ICAR is O(MN( E + V log V +D max )), were M is te maximum number of rounds. M is no larger tan 5 in our simulation. VII. EVALUATION We evaluate our real-time routing algoritms troug bot experiments on a pysical WSAN testbed and numerical simulations. As discussed in Section II, routing protocols designed for traditional wireless sensor networks are incompatible wit te WirelessHART standard, wereas recent efforts on WirelessHART routing ave focused on enancing te reliability [13] [15] and energy efficiency [18], [19] of multi-pat grap routing wic usually lead to larger latency tan single-pat routing like our routing algoritms. Hencefort, we compare our conflict-aware routing algoritms (CAR and ICAR) against Sortest Pat Routing (SP) as a baseline for performance evaluation. Note tat, wile link quality is often incorporated in routing metrics for traditional wireless sensor networks as a routing metric, it is not as useful for WirelessHART networks in wic links are usually igly reliable due to aggressive link blacklisting and conservative deployment. In suc networks, te sortest pat is a reasonable euristic to reduce latency. Comparing conflict-aware routing against sortest pat routing quantifies te benefit of considering transmission conflicts to real-time performance. Our evaluation includes two parts: (1) experiments on a WSAN testbed and (2) simulations based on network topology traces collected from pysical experiments. We evaluate routing protocols wit tree metrics: (a) Acceptance ratio: te percentage of test cases tat are deemed scedulable, (b) Endto-end delay: te communication delay between te release of a packet from te source and te reception at te destination, and (c) Execution time: te total time required to compute routes at te network manager. A. Experiments on a WSAN Testbed We evaluate our routing designs on an indoor WSAN testbed consisting of 63 TelosB motes. Figure 1 sows te topology of te WSAN testbed, and te locations of access points, sources and destinations of flows. For eac link in te testbed, we measure its packet reception ratio (PRR) by counting te number of received packets among 25 packets transmitted on te link. Following te practice of industrial deployment, we add only links wit PRRs iger tan 9% to te topology of te testbed. Te motes in te testbed run a

5 Access Point Source Destination RelayNode F1 F2 F3 F4 F5 F6 F7 F8 Delay (ms) Flow Priority Conflict Delay (ms) Flow Priority Fig. 1: WSAN testbed topology tat includes te locations of access points, sources, and destinations of flows. (a) End-to-end delays Fig. 2: Delays in experiments. (b) Conflict delays protocol stack [6] implementing source routing and TSCH on top of te CC242X radio stack. In our experiment, we generate 8 distinct flows, and use 8 cannels for communication. Due to cannel opping, eac transmission of any flow can op troug all cannels used. Te period of eac flow is in te range of milliseconds, wic are typical periods used in process industries, as specified in te WirelessHART standard [2]. Te lengt of te yper-period is 128 milliseconds. Te relative deadline of eac flow is equal to its period. We run te experiments long enoug tat eac flow can deliver at least 1 packets. We evaluate ow muc our approac improves end-to-end and conflict delays over te performances of SP. Figure 2(a) presents te worst-case end-to-end delays of eac flow. CAR and ICAR consistently acieve similar or better end-to-end delays tan SP. Moreover, ICAR can furter improve te delays of some lower priority flows compared to CAR. Note tat since te locations of a source and destination of a flow, and te network topology also impact te end-to-end delay of a flow, some of te lower priority flows may ave smaller delays tan tose of iger priority flows. We next investigate conflict delays of flows under our approaces and te baseline. Since conflict delay contributes to te end-to-end delay of a flow, reducing te conflict delay can enance te end-to-end delay. As sown in Figure2(b), under CAR and ICAR, flows incur less conflict delay tan under SP. For flow 2 and flow 7, CAR and ICAR can eliminate te conflict delay, wile SP still incurs some conflict delay. In addition, ICAR can outperform CAR in many cases since it uses an iterative algoritm tat allows iger priority flows to take into account routes of lower priority flows. B. Simulations based on te WSAN Testbed Topology To provide a more compreensive evaluation, we also evaluate our routing algoritms troug simulations based on te WSAN testbed topology. Te simulator uses te same routing and sceduling algoritms as in our testbed experiments and is written in C++. All simulations are performed on a MacBook Pro laptop wit 2.4 GHz Intel Core 2 Duo processor. We evaluate our algoritms under different numbers of cannels (from 4 to 15). Wit a given set of cannels, we test our routing designs on different numbers of flows by increasing te numbers of source and destination pairs from 2 to 22. Te period of te eac flow is randomly picked witin te range of milliseconds. Te relative deadline of eac flow is equal to its period. For te same number of flows, we run 1 tests wit randomly generated pairs of sources and destinations. In summary, we perform numerical evaluations on about 1K different configurations. Te following results are from simulations wit 8 cannels. Figure 3(a) compares te acceptance ratios of CAR, ICAR, and SP in simulations. SP always as te lowest acceptance ratio, and bot CAR and ICAR ave muc iger acceptance ratios. ICAR as a iger acceptance ratio tan CAR, wic sows te benefit of letting flows wit iger priorities be aware of te routes of lower priority flows. Compared to SP, CAR and ICAR can respectively improve te acceptance ratio by 239% and 35% on average. We ten compare te delays of CAR, ICAR, and SP under 8 cannels. As sown in Figure 3(c), conflict delay indeed dominates contention delay. Moreover, altoug CAR and ICAR may lead to routes wit longer op counts, teir endto-end delays are smaller tan SP on average as presented in Figure 3(b). Tis is because CAR and ICAR ave fewer conflict delays tan SP in all cases. We obtain similar results wit 4-7 and 9-15 cannels used. Witin all simulations, our CAR and ICAR algoritms improve te acceptance ratio significantly, wit smaller conflict delays. Wen te number of cannels is small (4-8), te contention delays can be an important part of te end-to-end delays. However, wen more (12-15) cannels are used, te contention delays are zero, and conflict delays dominate. Execution Time: We compare te execution time of SP, CAR, and ICAR in Figure 4. Te execution time increases as te number of flows increases in all tree algoritms. Te execution times of te tree routing algoritms follow te order SP<CAR<ICAR. SP as te lowest execution time since it uses te breadt-first searc algoritm. ICAR as a iger execution time tan CAR because it is an iterative algoritm. Te execution time of ICAR is less tan 2 ms wen te number of flows is 22. According to te WirelessHART Standard [2], wen a node loses connectivity to its neigbor for a certain period of time (timeout period), a node will send a keep-alive packet to probe te connection. Tis timeout period is no less tan 3 seconds. If te node fails to

6 Acceptance Ratio (%) Delay (ms) Delay (ms) SP-Conf CAR-Conf ICAR-Conf SP-Cont CAR-Cont ICAR-Cont (a) Acceptance ratios. (b) End-to-end delays (c) Conflict (Conf) and contention (Cont) delays. Fig. 3: Simulation results: acceptance ratio and delays. Execution Time (ms) SP CAR ICAR Fig. 4: Execution time receive a response from its neigbor, a node will issue a patdown alarm to te network manager, wic will recalculate a new route. Tis indicates tat te 2 ms execution time is acceptable for te real-world operations. VIII. CONCLUSIONS As process industries start to embrace WSANs, it becomes critical for WSANs to support real-time communication for IIoT. Due to limited results on real-time routing for industrial WSANs, tis paper proposes conflict-aware routing, a new approac to real-time routing in industrial WSANs tat considers transmission conflict delays in routing decisions. As a result, a WSAN can accommodate more real-time flows wile meeting teir deadlines. Evaluations based on bot testbed experiments and simulations sow tat conflict-aware routing can lead to up to tree-fold improvement in te real-time capacity of a WSAN wen compared to sortest pat routing. ACKNOWLEDGMENT Tis work is supported, in part, by te NSF troug grants (NeTS), (CPS), and (CRII). REFERENCES [1] C. Lu, A. Saifulla, B. Li, M. Sa, H. Gonzalez, D. Gunatilaka, C. Wu, L. Nie, and Y. Cen, Real-Time Wireless Sensor-Actuator Networks for Industrial Cyber-Pysical Systems, in Proceedings of te IEEE, Special Issue on Industrial Cyber Pysical Systems, vol. 14, pp , May 216. [2] WirelessHART Sfpecification, 27. ttp:// [3] ISA1: Wireless Systems for Automation. ttps:// isa1/. [4] IPv6 Over te TSCH Mode of IEEE e (6TiSCH). ttps://datatracker.ietf.org/wg/6tisc/about/. [5] A. Saifulla, Y. Xu, C. Lu, and Y. Cen, Real-Time Sceduling for WirelessHART Networks, in RTSS, 21. [6] M. Sa, D. Gunatilaka, C. Wu, and C. Lu, Empirical Study and Enancements of Industrial Wireless Sensor-Actuator Network Protocols, IEEE Internet of Tings Journal, vol. 4, pp , June 217. [7] RPL: IPv6 Routing Protocol for Low-Power and Lossy Networks. ttps://tools.ietf.org/tml/rfc655. [8] S. Duquennoy, O. Landsiedel, and T. Voigt, Let te Tree Bloom: Scalable Opportunistic Routing wit ORPL, in Sensys, 213. [9] T. He, J. A. Stankovic, C. Lu, and T. Abdelzaer, SPEED: A Stateless Protocol for Real-Time Communication in Sensor Networks, in ICDCS, 23. [1] O. Cipara, Z. He, G. Xing, Q. Cen, X. Wang, C. Lu, J. Stankovic, and T. Abdelzaer, Real-time Power-Aware Routing in Sensor Networks, in IEEE International Worksop on Quality of Service, 26. [11] J. Heo, J. Hong, and Y. Co, EARQ: Energy Aware Routing for Real-Time and Reliable Communication in Wireless Industrial Sensor Networks, IEEE Transactions on Industrial Informatics, vol. 5, pp. 3 11, Feb 29. [12] P. T. A. Quang and D. S. Kim, Enancing Real-Time Delivery of Gradient Routing for Industrial Wireless Sensor Networks, IEEE Transactions on Industrial Informatics, vol. 8, pp , Feb 212. [13] J. Zao, Z. Liang, and Y. Zao, ELHFR: A Grap Routing in Industrial Wireless Mes Network, in ICIA, 29. [14] S. Han, X. Zu, A. K. Mok, D. Cen, and M. Nixon, Reliable and Real-time Communication in Industrial Wireless Mes Networks, in RTAS, 211. [15] G. Gao, H. Zang, and L. Li, A Reliable Multipat Routing Strategy for WirelessHART Mes Networks Using Subgrap Routing, Journal of Computational Information Systems, vol. 9, Marc 213. [16] J. Niu, L. Ceng, Y. Gu, L. Su, and S. K. Das, R3E: Reliable Reactive Routing Enancement for Wireless Sensor Networks, IEEE Transactions on Industrial Informatics, vol. 1, pp , Feb 214. [17] L. Pradittasnee, S. Camtepe, and Y. C. Tian, Efficient Route Update and Maintenance for Reliable Routing in Large-Scale Sensor Networks, IEEE Transactions on Industrial Informatics, vol. 13, pp , Feb 217. [18] C. Wu, D. Gunatilaka, A. Saifulla, M. Sa, P. B. Tiwari, C. Lu, and Y. Cen, Maximizing Network Lifetime of WirelessHART Networks under Grap Routing, in IoTDI, 216. [19] S. Zang, A. Yan, and T. Ma, An Energy-Balanced Grap Routing Algoritm for WirelessHART Networks, in IHMSC, 213. [2] S. Duquennoy, B. Al Naas, O. Landsiedel, and T. Watteyne, Orcestra: Robust Mes Networks Troug Autonomously Sceduled TSCH, in Sensys, 215. [21] Y. Xu, F. Ren, T. He, C. Lin, C. Cen, and S. K. Das, Real-time Routing in Wireless Sensor Networks: A Potential Field Approac, ACM Transactions on Sensor Networks, pp. 35:1 35:24, June 213. [22] S. M. S. Nirjon, J. A. Stankovic, and K. Witeouse, IAA: Interference Aware Anticipatory Algoritm for Sceduling and Routing Periodic Real-time Streams in Wireless Sensor Networks, in INSS, 21. [23] A. Saifulla, Y. Xu, C. Lu, and Y. Cen, End-to-End Delay Analysis for Fixed Priority Sceduling in WirelessHART Networks, in RTAS, 211. [24] C. Wu, M. Sa, D. Gunatilaka, A. Saifulla, C. Lu, and Y. Cen, Analysis of EDF Sceduling for Wireless Sensor-Actuator Networks, in IWQoS, 214.