Topology Construction in RPL Networks over Beacon-Enabled

Transcription

1 Toology Construction in RPL Networks over Beacon-Enabled 8.. Mališa Vučinić, Gabriele Romaniello, Laurène Guelorget, Bernard Tourancheau, Franck Rousseau, Olivier Alhand, Andrzej Duda, and Laurent Damon Grenoble Als University, CNRS Grenoble Informatics Laboratory UMR 7, France STMicroelectronics, Crolles, France to construct a cluster-tree anchored at the PAN coordinator (also the sink node) for suorting multi-ho communication. Moreover, a node joining the cluster-tree has to associate with a coordinator (a Layer oeration) before it may send any data frame. The choice of the coordinator influences any ossible choice of the RPL arent node. In the case of the beacon-enabled IEEE 8.. nodes, the roblem is how to construct the 8.. cluster-tree according to the RPL routing information based on a DODAG. While both beacon-enabled IEEE 8.. and RPL have been extensively studied within their OSI abstraction layer, the joint oeration is surrisingly still an oen roblem. The existing work in the literature [] requires extensive modifications to both standards, which is an unrealistic requirement at the current stage of IoT stack develoment. We roose a solution to the roblem that satisfies the constraint of keeing RPL and IEEE 8.. unchanged. In our aroach, RPL constructs its DODAG before the clustertree at L and we use the RPL routing information (selection of the referred arent) in the association decision to establish links, i.e., to select the coordinator in the cluster-tree that is the referred arent in the DODAG. The roosed solution takes advantage of cross-layer signaling: a node joining the network requests RPL information from neighbor 8.. coordinators and associates with the right coordinator based on the information in the RPL DIO message (DODAG Information Object). We adat the oeration of the Trickle timer [6] that governs the transmission of DIO messages to rovide the required information to Layer (the adatation remains comliant with the RPL secification). The main contributions of the aer are the following: a new scheme that allows RPL to run over the beaconenabled IEEE 8.. without any modification to the two standards, the scheme leading to energy savings both during the toology construction and in the steady-state, due to the use of the Trickle timer, a simle robabilistic model of the Trickle timer and an analysis of the delay of the roosed scheme, an evaluation of energy savings and the time for toology convergence based on the imlementation of the roosed scheme in Contiki. The aer is organized as follows. We resent the backarxiv:.78v [cs.ni] Ar Abstract In this aer, we roose a new scheme that allows couling beacon-enabled IEEE 8.. with the RPL routing rotocol while keeing full comliance with both standards. We rovide a means for RPL to ass the routing information to Layer before the 8.. toology is created by encasulating RPL DIO messages in beacon frames. The scheme takes advantage of 8.. command frames to solicit RPL DIO messages. The effect of the command frames is to reset the Trickle timer that governs sending DIO messages. We rovide a detailed analysis of the overhead incurred by the roosed scheme to understand toology construction costs. We have evaluated the scheme using Contiki and the instruction-level Cooja simulator and comared our results against the most common scheme used for dissemination of the uer-layer information in beaconenabled PANs. The results show energy savings during the toology construction hase and in the steady state. Index Terms IEEE 8.., beacon-enabled mode, RPL, Wireless Sensor Networks, toology construction, multi-ho networks. I. INTRODUCTION The long awaited Internet of Things (IoT) has never been closer. The industry has fully begun to take art and the further develoment is all about standard comliance. The uer layers of the IP rotocol stack for Wireless Sensor Networks (WSN) are being fine-shaed and the gas between IETF and IEEE standards [] are being bridged. Full solutions begin to emerge so roduct interoerability and security are of rimary concern []. The IEEE 8.. standard [] is widely recognized as the main technology for low-ower wireless sensor networking []. Among its different modes, we focus on the beaconenabled mode able to achieve very low energy consumtion by suorting a desired level of radio duty cycling (the roortion between the eriods nodes are on and off). In this aer, we address the roblem of running the Routing Protocol for Low Power and Lossy Networks (RPL) [], the IETF standard for routing in WSN, on to of IEEE 8.. beacon-enabled nodes. The forwarding structure built by RPL is a Destination Oriented Directed Acyclic Grah (DODAG). Each node kees a list of available arent nodes closer to the DODAG root (sink node) and selects one of them as the referred arent based on an objective function. When a link to the referred arent fails, a node switches to another arent in its list. At the link layer, the beacon-enabled IEEE 8.. nodes need

2 ground information on the beacon-enabled IEEE 8.. and RPL in Sections II and III. We rovide a detailed descrition of the roosed scheme in Section IV and evaluate it in Section V. Section VI summarizes the related work. We conclude and discuss the future work in Section VII. II. BEACON-ENABLED IEEE 8.. The lack of a radio duty cycling scheme in the IEEE 8.. non-beacon mode has led to intensive research on the beacon-enabled mode. Periodic beacon frames allow the synchronous slee schedule of devices in the network. As a consequence, it is ossible to guarantee a desired level of radio duty cycling (RDC), which is of the utmost imortance for battery oerated nodes and esecially for those that harvest energy from the environment. The oeration of nodes in the beacon-enabled mode relies on beacons that delimit the start of a suerframe. Immediately following is the Contention Active Period (CAP) during which nodes transmit ending data frames to their arent (cluster coordinator) using the slotted CSMA/CA algorithm (a coordinator node needs to stay active during CAP). Beacon Order (BO) and Suerframe Order (SO) are the key arameters to tune the desired level of radio duty cycling in the beacon-enabled mode. Beacon Interval () is defined as = abasesuerf rameduration BO and SD = abasesuerf rameduration SO is the CAP duration. Thus, the uer-bound roortion of time a duty cycled coordinator node will be active is SO BO. Leaf nodes that only wake u for a transmission may benefit from an even lower duty cycle. The network formed in the non-beacon mode may be a mesh in which each node may communicate with its radiorange neighbors, so running RPL in this case does not raise any roblems. Nodes in the beacon-enabled mode have to form a cluster-tree: a node selects one arent node, the cluster coordinator, and synchronizes with its beacons. The node may become a coordinator itself on behalf of other nodes, which enables multi-ho communication from leaf nodes to the root of the cluster-tree. Scheduling of active eriods of different coordinators is not defined in the IEEE 8.. standard and is therefore left as an imlementation choice. In our evaluations, we use a simle static allocation of active eriods that allows us to focus on the toology construction without affecting the overall results. The Personal Area Network (PAN) coordinator is the root of the tree, the sink of the sensor network. It starts the toology construction by transmitting the first beacon. Other nodes are unassociated and have to switch their radio transceivers on to erform assive scanning, the only mechanism for discovering otential coordinators available in the beaconenabled mode. The recetion of a beacon initiates a scan eriod during which a node waits for beacons. At the end of this eriod, a node can initiate the association with the best coordinator with the sequence of association-request, ack, data-request, association-rely, ack control frames. Note that most of the energy consumed during the toology construction hase comes from idle listening during the scan eriod, which is unavoidable for any association strategy that discovers the best available coordinator. The duration of this interval should allow the discovery of all coordinators in the radio range. Fig. illustrates a timeline of the toology construction for an examle cluster-tree comosed of four nodes. Note that Node may receive beacons from Coordinators and, but it selects Node as the best arent. Scan eriod Association request ACK Beacon Tx Association resonse Data request Fig. : Toology construction in an examle 8.. clustertree. III. RPL ROUTING PROTOCOL FOR LOW POWER AND LOSSY NETWORKS Oerating at the Network layer, RPL [] is a Distance Vector rotocol that secifies how to construct a Destination Oriented Directed Acyclic Grah (DODAG) with a defined objective function and a set of metrics and constraints. RPL secifies a set of new ICMPv6 control messages to exchange information related to a DODAG and notably: DODAG Information Object (DIO) defines and maintains uward routes to the root, i.e. the DODAG. DODAG Information Solicitation (DIS) messages request the DODAG related information from neighboring nodes. The root starts the DODAG building rocess by transmitting a DIO. Neighboring nodes rocess DIOs and make a decision on joining the DODAG based on the objective function and/or local olicy. A node comutes its Rank with resect to the root and starts advertising DIO messages to its neighbors with the udated information. As the rocess continues, each node in the network receives one or more DIO messages and selects a referred arent towards the root. Note that for robustness, RPL kees a list of other arents that can be used in case link conditions change. As we focus on toology construction, we limit the discussion to uward routes. In case of beacon-enabled IEEE 8.. at L, the traditional layer-indeendent oeration would confine the selection of RPL routes to those in the already-constructed L clustertree. Consequently, the overall erformance of RPL would be significantly degraded. We exloit the aroach of merging two structures: the 8.. cluster-tree and the DODAG of t

3 RPL, which allows us to benefit from low overhead, small delays, and near otimal uward routes of RPL [7] while creating the IEEE 8.. cluster-tree required for low duty cycle communications. RPL uses the Trickle algorithm [6] to govern the transmission interval of DIOs. The Trickle algorithm has three main arameters: i) the minimum interval size I min, ii) the maximum interval size I max exressed as the number of times the timer may double, iii) the redundancy constant k. The main idea of the algorithm is to exonentially reduce the amount of control traffic in the network when the toology is consistent, i.e. when there are no link failures or arriving nodes. Consistency is checked by comaring the DODAG state advertised by other nodes in DIOs with the local one. If the number of consistent DIO recetions is higher than redundancy constant k, Trickle refrains from transmitting. Instant t at which Trickle decides if it is going to transmit is randomly selected from interval [I/, I), where I {I min n n Z, n I max }. Interval I is doubled uon its exiration by incrementing n. When a node detects inconsistency (which also includes the initial DODAG construction), n becomes, which sets interval I to I min. As detailed in the next section, we suitably adat arameter I min to construct IEEE 8.. cluster-tree according to the DODAG of RPL. IV. 8.. CLUSTER-TREE CONSTRUCTION BASED ON RPL DODAG We roose the selection of the best coordinator in the 8.. cluster-tree based on the referred arent in the DODAG of RPL. The resulting cluster-tree will effectively be a subset of the DODAG initialized during the toology construction hase. There are several issues with such an aroach: ) RPL is a network layer rotocol, but no communication among nodes at Layer may take lace before links at Layer are established (node association with a coordinator). ) An IEEE 8.. node once associated can only communicate with its cluster coordinator, so after association, a node can only receive DIO messages from its cluster coordinator. In this aer, we focus on addressing the first issue to enable toology construction. The second issue is a art of the future work that may add robustness to IEEE 8.. cluster-tree by keeing alternate arents from the DODAG of RPL. To address the first issue, we exloit the fact that DIO messages are multicast. As Layer multicasts translate to Layer broadcasts, we use beacons to broadcast DIO messages. There is no better broadcast mechanism in multi-ho beacon-enabled networks than the beacons themselves during the scan eriod devices wait for beacons. We assume that IEEE 8.. Reduced Function Devices (RFD) are configured as RPL leaf nodes, i.e., they do not send DIO messages. Similarly, Full Function Devices (FFD) may become cluster coordinators, i.e., they have to be configured as RPL routers, which is a realistic assumtion as the role of a device mainly deends on its energy source. We assume that a node a riori knows if it is an RFD or an FFD. We roose the encasulation of RPL DIO messages in the beacon frame ayload following an idea discussed in the team [8]. Layer adds DIO to the ayload of the next scheduled beacon if the resulting frame does not exceed IEEE 8.. maximum hysical layer frame size of 7 bytes (cf. Fig. ). In case the DIO message cannot fit into the current beacon, it may be fragmented or delayed for the following one as the beacon ayload size varies as a function of downward traffic. RPL IP 6LoWPAN IEEE 8.. DIO event Beacon Tx DIO event DIO ayload Fig. : Encasulation of DIO messages in beacon frames. The exonential increase of the DIO transmission interval governed by Trickle has an imortant side effect: arriving nodes would otentially wait a long time interval before receiving the first DIO. RPL addresses this issue with DIS messages that can be broadcast to solicit the transmission of a DIO: uon recetion of a DIS, a node resets its Trickle interval I to I min so DIO will be transmitted shortly []. However, DIS broadcast is not enough for synchronous duty cycled networks neighbor nodes in the radio range may slee at the instant of the DIS transmission. As exlained above, the recetion of a beacon delimits the start of the Contention Active Period during which the coordinator is active. Thus, CAP is the most suitable eriod during which an unassociated node may solicit information from nearby coordinators. Note that a node wanting to join the network is awake during the scanning eriod so it can receive beacons from several neighbor coordinators. Thus, we roose that the node transmits a solicitation message by erforming CSMA/CA after the beacon if the following two conditions hold: the received beacon is the first one received from a given coordinator, the beacon does not contain a DIO in its ayload. The solicitation message could be a RPL DIS message encasulated in an 8.. command frame. Note that a node cannot send data frames before association []. However, we have chosen to use the IEEE 8.. beacon-request command frame without any ayload as a solicitation message it has a small size (8 bytes) so a very short transmission time. Additionally, the RPL secification [] allows the Trickle reset triggered by external events. Note that the beacon-request command frame is tyically used in the non-beacon mode to solicit the information about the network. It has no use in the beacon-enabled mode as beacons are eriodically transmitted. We use its recetion at Layer to trigger the reset of the Trickle timer at the RPL layer t

4 to sawn a DIO transmission. The goal is to encasulate the DIO message in the following beacon so that arriving nodes can select the best coordinator. As a node may send several beacon-request solicitation frames during the scan eriod (and CAP of each detected coordinator), the scheme ensures the reset of the Trickle timer for all RPL routers in the range. A ossible drawback of the scheme could be its ossible side effect on the duration of the always-on scan eriod. In fact, with tyical arent selection schemes at Layer, each beacon carries a network-secific metric rocessed by arriving nodes. Then, in case Beacon Order is a riori known, the worst-case scan duration is one Beacon Interval. However, a simle algorithm achieves the same duration with our scheme as well during the scan eriod of duration : ) for each discovered coordinator, a node stores the exected instant of the next beacon (current time() + ), ) for each discovered coordinator, a node solicits the reset of the Trickle timer as exlained above, ) uon exiration of, a node goes to slee and schedules its wake u at the instants found in (), ) a node wakes u and receives the beacon with the DIO ayload, ) uon recetion of the DIO ayload from the last discovered coordinator, a node consults RPL about the best choice and schedules the next wake u just before the beacon of the selected coordinator; then, the node follows the standard association rocedure. This scheme ensures the discovery of all coordinators in the radio range while allowing a node to start duty cycling after one from the boot time (cf. Fig. ). During next, node receives DIOs and asses them to RPL. In the worst case, by the end of the second, RPL will have the referred arent selected. The additional worst-case delay of one is the rice to ay during the toology construction for the benefit that comes later-on with the Trickle timer during the network oeration. As the node sends most of the second beacon interval sleeing, it consumes energy only for receiving beacons. For n discovered coordinators, the energy will be E = n T I RX V, where I RX is the radio current draw in receive mode, V the oerating voltage, and T transmission time of one IEEE 8.. beacon with a DIO message in its ayload (tyically around. ms for kb/s IEEE 8.. comliant radios). Note, however, that in many deloyments, BO is not a- riori known. In such cases, devices have to scan for longer eriods to account for the largest exected in resence of multile PANs [9]. Our scheme in such scenarios introduces no additional delay as long as the reconfigured scan duration is greater than or equal to half the actual in the network. A. I min Parameter Tuning and Analysis The successful oeration of the roosed scheme requires that, uon solicitation, the subsequent beacon includes a DIO message. To achieve such behavior while keeing the oeration of two layers indeendent, we need to configure beginning of scan CAP scan eriod Association request Association resonse sleeing eriod Beacon Tx DIO ayload sleeing eriod DIO Solicitation frame Data request ACK Fig. : Soliciting DIO during the scan eriod. t Trickle reset the Trickle I min arameter as a function of, because the recetion of a solicitation frame triggers the Trickle timer reset and the next timer value will be uniformly drawn from the interval [I min /, I min ). Thus, to ensure the arrival of the next DIO before the subsequent beacon, the following condition needs to hold: I min SD, () where SD denotes CAP duration. Similarly, as reviously discussed, the worst case scan eriod when BO is a riori known, is. The otimal erformance of Trickle with our scheme is obtained when I min = SD, which ensures the successful oeration while having the lowest overhead. B. Analysis of DIO Recetion Delay We evaluate here the exected delay of DIO messages encasulated in eriodic beacons. We define the Trickle timer value as random variable X uniformly distributed in [I/, I), where I is a random variable denoting the current Trickle state. Then, from the Layer oint of view, a DIO message arrives during a beacon interval at instant X mod. Delay D is the interval remaining until the transmission of the next beacon: D = (X X ). () The exected delay is then: X E[D] = E[X] + E[ ]. () Now, recall that I is a discrete random variable in {I min n }, where n =,,..., I max. We model I with a discrete-time Markov chain shown in Fig., where denotes the robability of the Trickle reset. We can notice from I I I I... Imax Fig. : Markov chain with I max + states for Trickle. Fig. that stationary robabilities of states I,..., I Imax

5 follow a geometric distribution with reset robability : Π Ii = ( ) i, i =,..., I max. The last state, I Imax has the stationary robability: Π IImax = ( ) Imax. We can find the exected Trickle timer value as E[X] = E[E[X I]]. As our scheme uses the beacon-request solicitation frame at L to reset Trickle, the case I = I min is of a articular interest. From Eq., it follows that: X E[D I=Imin ] = E[X I=Imin ] + E[ I=Imin ]. () Given the condition of Eq. and also the fact that the right endoint is excluded from the uniform interval, term E[ X ] goes to zero leaving: E[D I = I min ] = E[X I = I min ]. Finally, as X is now a uniform random variable in [I min /, I min ), the exected DIO delay becomes: E[D I = I min ] = I min, I min. () We have validated Eq. by emulating a real node running the Contiki oerating system for constrained devices. We have timestamed the exiration instants of Trickle and the instants of the beacon with DIO transmission. We have configured I min to an aroximate value of / (Contiki accets the values of I min in ower of ). The emulation results over samles strongly corroborate our analysis with a maximal error of.799%. From Eqs. and, it follows that for setting I min = SD, our scheme introduces the least additional delay to Trickle after reset, while ensuring successful oeration Pan coordinator Coordinator 6 Device 6 Cluster Fig. : Evaluated toology. V. PERFORMANCE EVALUATION To evaluate our scheme, we have used an imlementation of the IEEE 8.. beacon-enabled mode secifically develoed and otimized for harvested sensor motes manufactured by STMicroelectronics (ST) containing a bit microcontroller and a rorietary radio 8.. transceiver. To our knowledge, it is the first IEEE 8.. beaconenabled imlementation for the Contiki oerating system. To benefit from the Cooja simulator [] that uses the MSPsim instruction-level emulator of the Tmote Sky latform, our team orted the beacon-enabled layer develoed for the ST motes to the Tmote Sky latform. Tmote Sky is based on a 6-bit MSP microcontroller, oerating at 8MHz clock rate, and the CC Chicon radio. Note that the only imerfection of Cooja with resect to the real world environment comes from the Unit Disk Grah radio channel model. Fig. resents the evaluated toology. Many authors in the literature discussed the method of encasulating information necessary for toology construction in the beacon ayload (arent selection, neighbor discovery) [], [], []. Consequently, they assume the information to be resent in each beacon. As our goal in this aer was to resent benefits in terms of 8.. toology construction, we have comared our scheme against this aroach and denote the scheme Systematic Beacon Payload (SBP). To be fair and not to loose the generality of our results, we have studied the effects of varying the SBP message size and how it affects erformance. We found that the two schemes have similar erformance when the SBP message size is aroximately / of the DIO size (cf. Fig. 6(a)), that is, when one coordinator from Fig. sends DIO message for every beacons with SBP on the average during toology construction. Note that this ratio deends on the duration of the scan eriod and the configuration of Trickle. For a given imlementation, one can easily evaluate such a ratio and derive the gain or loss deending on the message size arameters. We set the I min Trickle arameter to aroximately SD and kee SO equal to. We comute the radio energy consumtion from the current draw values reorted in the Tmote Sky data sheet. We average all the oints in the following grahs over emulation runs and show them with 9% confidence intervals. We can notice in Fig. that nodes have only one coordinator in their radio range. We have chosen such toology to focus on toology construction in RPL networks over beaconenabled 8.. and evaluate the effect of our scheme. In this way, we isolate toology construction asects from the roblems related to routing that may deend on the choice of routing metrics or objective functions. Moreover, a single coordinator discovered during the scan eriod means that the solicitation scheme is ut under stress. Indeed, if a single DIO message does not arrive with the subsequent beacon uon solicitation, the node will have to initiate another scan eriod, which would unnecessarily increase the toology convergence delay. Nevertheless, the examle toology in Fig. is favorable to the roosed scheme in terms of delay it does not introduce additional delay in case BO is a riori known, i.e., the first discovered coordinator is also the last one, so a node can initiate the association rocedure after the scan eriod of one. However, we discuss the worst case delay in the resence of multile coordinators in Section IV. Also note that in some cases, the first beacon discovered during the scan eriod may already contain a DIO message. As the Trickle timer randomly selects its exiration interval and our scheme kees the oeration of two layers indeendent, it is a lucky outcome. In this case, a node does not need to solicit DIO as detailed in Section IV. However, a node still has

6 Control overhead [Kbytes] Energy consumtion in TX [mj] SBP BO= DIO BO= SBP BO=6 DIO BO=6 /6 / / / /6 SBP/DIO ratio (a) Overhead, variable SBP size... SBP ho SBP ho SBP ho DIO ho DIO ho DIO ho Association delays [s] SBP ho SBP ho SBP ho DIO ho DIO ho DIO ho (b) Toology convergence time. (c) Energy sent in transmission, in mj. (d) Energy sent in recetion, in J. Energy consumtion in RX [J].. SBP ho SBP ho SBP ho DIO ho DIO ho DIO ho Fig. 6: Results from emulation during toology construction. to wait for the exiration of the scan eriod before initiating its association rocedure to ensure that it has discovered all otential coordinators. Contol Overhead [Kbytes] 6 SBP DIO 8 Scan Duration (a) Total overhead Energy consumtion in TX [mj]... SBP ho SBP ho SBP ho DIO ho DIO ho DIO ho 8 Scan Duration (b) Energy sent in transmission, in mj Fig. 7: Results from emulation during toology construction for variable scan duration and BO =. We resent the results for the case in which two schemes have the most similar erformance, i.e., we set the message size of SBP to / of DIO (cf. Fig. 6(a)). Larger SBP message sizes result in worse erformance while smaller SBP messages result in better erformance during toology construction in case BO is a riori known. A. Toology Construction We study the toology construction hase for two cases: ) BO is a riori known so the scan eriod can be set to the minimal value of ; ) there is no a riori knowledge of BO so nodes use a sub-otimal scan duration to account for the worst case. In both cases, simulations last until the association of the last node. For case ), Figs. 6(b)-6(d) resent the results for varying BO. We can see in Fig. 6(b) that our scheme does not introduce any additional delay for the evaluated toology and the results for two schemes are similar within confidence intervals. Fig. 6(c) shows similar results in terms of cumulative energy sent in transmission, a consequence of the choice of the arameters for two schemes. Notably, coordinators at ho and send aroximately the same energy transmitting beacons. The major art of the energy sent in recetion comes from idle listening during the scan eriod so two schemes erform equally (cf. Fig. 6(d)). For case ), when BO is not a riori known, we vary the scan eriod. As nodes remain in recetion mode much longer, the energy sent in recetion makes the major art of the total consumtion. Similarly to Figs. 6(b) and 6(d), two schemes erform equally. However, as the scan eriod is longer, there is a larger number of beacons transmitted before the toology converges. We can thus see the effect of the Trickle algorithm and the roosed solicitation scheme (cf. Fig. 7(a)) that results in energy savings for ho nodes as they transmit beacons the longest until the end of the tree construction (cf. Fig. 7(b)). B. Steady-state Furthermore, we have evaluated the benefits in terms of energy savings in the steady state, i.e., after toology construction. There was no alication traffic in the network and nodes simly duty cycle according to their schedules. The resented results concern 6 minutes of the network oeration after the association of the last node. We can see the effect of the reduction in control overhead by the Trickle algorithm in Fig. 8(a). In articular, FFD nodes (ho and ) transmit short beacons without any ayload most of the time, which results in energy savings both during recetion and transmission. During recetion, however, a major art of energy consumtion comes from active listening during the CAP of each coordinator so this effect is masked (cf. Fig. 8(b)). Note that in Fig. 8(a), the consumtion of RFDs is zero as there is no alication traffic in the network. Also, during the steady state, the recetion consumtion of FFDs (ho and ) is the same, as devices remain active during the same amount of time (CAP duration). Energy consumtion in TX [mj] 7 6 SBP ho SBP ho DIO ho DIO ho (a) Energy sent in transmission, in mj. (b) Energy sent in recetion, in J. Energy consumtion in RX [J] SBP ho SBP ho SBP ho DIO ho DIO ho DIO ho Fig. 8: Results from emulation during 6 min. of steady state. VI. RELATED WORK The erformance of multi-ho IEEE 8.. networks has been well studied during the last years both using robabilistic aroaches [] and simulations []. Energy consumtion introduced during the scan eriod is widely recognized as a

7 significant roblem. The recent work of Karowski et al. [9] lowered this cost by otimizing the number of slots to listen over different channels. Romaniello et al. [6] roosed the Multichannel Beacon Train Protocol for faster discovery over multile channels in the resence of varying beacon intervals. Kohvakka et al. discussed a rotocol that carries the time offset and the frequency channel in beacons to ease the scanning rocess for the joining node []. It is imortant to stress that our work in this aer is agnostic of the scanning rocess. Namely, the solicitation scheme we roose starts once a node has discovered all neighboring coordinators. As the de-facto standard for routing in IP-based WSN, RPL has been extensively studied in terms of convergence delays, route otimality, ath availability, and incurred overheads [7], [8]. Couled with the common wisdom that cross-layer signaling is necessary for a successful oeration of a routing rotocol in low ower and lossy networks, this fact rovides a strong suort to the aroach resented in our aer. The work of Pavković et al. is closely related to ours []. The authors roosed the adatations to the IEEE 8.. standard to integrate its oeration with RPL. Moreover, they roosed an oortunistic version of RPL to imrove the delivery of time-sensitive traffic and evaluated the roosal in terms of acket delivery ratio and delay. In recent work [], they discussed the RPL erformance benefits of modifying the IEEE 8.. cluster-tree structure into a cluster-dag. Our work was basically motivated by the same roblem the incomatibility of two structures, the 8.. clustertree and the DODAG. While the aroach of Pavković et al. resents erformance imrovement, its main drawback is the need for modifications of two standards, RPL and IEEE 8... We have addressed the same roblem from a different ersective instead of modifying the standards, we rovide a means for constructing the RPL DODAG and forming the cluster-tree as its subset. As a consequence, we obtain full comliance with both standards. VII. CONCLUSION We have resented a scheme that allows couling beaconenabled IEEE 8.. with the RPL routing rotocol. The scheme does not require any modification to both standards. We rovide a means for RPL to ass the routing information to Layer before the 8.. toology is created by encasulating RPL DIO messages in beacon frames. The scheme takes advantage of 8.. command frames to solicit DIO messages. The effect of the command frames is to reset the Trickle timer that governs sending of DIO messages. We have evaluated the roosed scheme using the Contiki oerating system for constrained nodes and the instructionlevel Cooja simulator. The results show energy savings during the toology construction hase and in the steady state. To isolate evaluation, we have disabled the RPL suort for downward traffic (DAO messages). We lan to evaluate its behavior when running on to of the IEEE 8.. clustertree as a art of the future work. We also lan to consider an otimal L mechanism that will allow the selection of an alternate coordinator from the DODAG in case of link failures. ACKNOWLEDGMENTS Many thanks to Jean-Batiste Guet and Chi-Anh La for the develoment of tools that we used during this work, and to Martin Heusse and anonymous reviewers for comments that imroved the quality of the aer. The work of O. Alhand, F. Rousseau, and A. Duda was artially suorted by the French National Research Agency (ANR) roject IRIS under contract ANR--INFR-6 and the Euroean Commission FP7 roject CALIPSO under contract The work reflects only the authors views; the Euroean Community is not liable for any use that may be made of the information contained herein. REFERENCES [] IEEE Std 8..- (Revision of 6),.,. [] M. Vučinić, B. Tourancheau, F. Rousseau, A. Duda, L. Damon, and R. Guizzetti, OSCAR: Object Security Architecture for the Internet of Things, in WoWMoM. IEEE,. [] M. Palattella, N. Accettura, X. Vilajosana, T. Watteyne, L. Grieco, G. Boggia, and M. Dohler, Standardized Protocol Stack for the Internet of (Imortant) Things, Communications Surveys Tutorials, IEEE, vol., no.,. 89 6,. [] T. Winter, P. Thubert, A. Brandt, J. Hui, R. Kelsey, P. Levis, K. Pister, R. Struik, J. Vasseur, and R. Alexander, RPL: IPv6 Routing Protocol for Low ower and Lossy Networks, IETF, RFC 6, March. [] B. Pavković, A. Duda, W.-J. Hwang, and F. Theoleyre, Efficient Toology Construction for RPL over IEEE 8.. in Wireless Sensor Networks, Ad Hoc Networks, Elsevier,. [6] P. Levis, N. Patel, D. Culler, and S. Shenker, Trickle: a Self-Regulating Algorithm for Code Proagation and Maintenance in Wireless Sensor Networks, in NSDI, vol.. USENIX Association,,.. [7] A. Triathi, J. de Oliveira, and J. Vasseur, Performance Evaluation of Routing Protocol for Low Power and Lossy Networks, IETF, RFC 6687,. [8] O. Alhand, E. Duble, A. Duda, M. Favre, R. Guizzetti, M. Heusse, and F. Rousseau, GreenNet: a Wireless Sensor Network for Harvested Nodes, Grenoble Informatics Laboratory, Tech. Re.,. [9] N. Karowski, A. Viana, and A. Wolisz, Otimized Asynchronous Multichannel Discovery of IEEE 8..-Based Wireless Personal Area Networks, Transactions on Mobile Comuting, IEEE, vol., no., ,. [] F. Osterlind, A. Dunkels, J. Eriksson, N. Finne, and T. Voigt, Cross- Level Sensor Network Simulation with COOJA, in Local Comuter Networks. IEEE, 6, [] M. Kohvakka, J. Suhonen, M. Kuorilehto, V. Kaseva, M. Hännikäinen, and T. D. Hämäläinen, Energy-Efficient Neighbor Discovery Protocol for Mobile Wireless Sensor Networks, Ad Hoc Networks, Elsevier, vol. 7, no.,., 9. [] B. Pavković, F. Theoleyre, and A. Duda, Multiath Oortunistic RPL Routing over IEEE 8.., in Proceedings of the MSWiM Conference. ACM,, [] Z. Yiming, Y. Xianglong, G. Xishan, Z. Mingang, and W. Liren, A Design of Greenhouse Monitoring Control System Based on ZigBee Wireless Sensor Network, in WiCom, 7, [] J. Misic, S. Shafi, and V. Misic, Performance of a beacon enabled IEEE 8.. cluster with downlink and ulink traffic, Parallel and Distributed Systems, IEEE Transactions on, vol. 7, no., 6. [] G. Anastasi, M. Conti, M. Di Francesco, and V. Neri, Reliability and energy efficiency in multi-ho ieee 8../zigbee wireless sensor networks, in ISCC. IEEE,,. 6. [6] G. Romaniello, E. Potetsianakis, O. Alhand, R. Guizzetti, and A. Duda, Fast and Energy-Efficient Toology Construction in Multi-Ho Multi- Channel 8.. Networks, in WiMob. IEEE,. [7] M. Vučinić, B. Tourancheau, and A. Duda, Performance Comarison of the RPL and LOADng Routing Protocols in a Home Automation Scenario, in WCNC. IEEE, Aril, [8] H. Kermajani and C. Gomez, Route change latency in low-ower and lossy wireless networks using rl and 6lowan neighbor discovery, in ISCC. IEEE,,