[IEEE 2008 28th International Conference on Distributed Computing Systems Workshops (ICDCS...

6
Improving routing reliability on wireless sensors network with emergency paths Bastien Mainaud, Mariem Zekri and Hossam Afifi Wireless Networks and Multimedia Services Department of the National Institute of Telecommunications, 9, Charles Fourier - 91011 Evry Paris France {bastien.mainaud,mariem.zekri,hossam.afifi}@it-sudparis.eu Abstract One of the most important issues in Wireless Sensor Networks is reliable communication. In this article, we present MAODV-SIM, a routing protocol that increases the reliability in a sensor network and which is based on AODV. Our protocol introduces multipath and signal intensity metric. We need to define what reliability is? How can we say that a protocol is more reliable than another? Which benchmark can be used? Multipath is needed to determine emergency routes that are required to find a new path to the destination if the first one fails. Then,to choose more reliable routes we use a different metric, the signal intensity metric which is based on 802.15.4 standard that defines two physical measurements : The Link Quality Indication(LQI) and the Energy Detection (ED). Simulations show that MAODV-SIM increases reliability in sensor network. Index Terms - Wireless Sensor Network, Routing Protocol, Reliabilty, Multipath 1 Introduction The need for reliable and robust communications is strong in WSN [1]. In case of medical application, com- munications should not be interrupted or corrupted. Sensor networks are considered to be the future mea- suring devices making the monitoring more comfort- able for the patient,scalable for the medical staff and cost effective [11]. Reliability can be studied on dif- ferent layers of the OSI model, in our case, we choose to work on the low layers, especially the routing layer. In a classical wired network, the physical environment does not generally interfere with data transmission. This is not true in a WSN where Sensor nodes could be out of order due to the nature, and a link between two nodes can simply disappear. As the network uses a hop-by-hop communication, node failure may lead to global network failure. The need for reliability is hence quite important. The goal of our solution is not to defined an efficient routing protocol, a shortest path protocol, neither a low consumption routing protocol, but a reliable routing protocol. We develop a solu- tion using multipath and strength intensity message metrics. Moreover, the solution we present shows also good performance in the other criteria. The paper is organized as follows. Section 2 presents the notion of reliability and some related works which deals with reliable protocols. Section 3 describes our solution. Section 4 presents some experimental results. Conclusion is presented in section 5. 2 Background 2.1 Reliability Given the nature of the application, it is absolutely critical that a query update reach the sensors in a re- liable manner. Any sensor network that is deployed to cater to a critical application, in both civilian and military environments, will require mechanisms to en- sure reliable delivery of information from the sink to the sensors [12]. First, we need to differentiate the reliability of a WSN node and the reliability of the entire WSN. The first case is the capacity of a node to be operational as long as possible. The constraints are multiple. First the downloaded software on the node must be safe and clean of error (debugging is not possible on a patient). Second, the power consumption is also strongly im- portant : a code optimisation effort must be done for reducing the power consumption and so increasing the node lifetime. A node which is off is a node that cannot The 28th International Conference on Distributed Computing Systems Workshops 1545-0678/08 $25.00 © 2008 IEEE DOI 10.1109/ICDCS.Workshops.2008.66 545

Transcript of [IEEE 2008 28th International Conference on Distributed Computing Systems Workshops (ICDCS...

Page 1: [IEEE 2008 28th International Conference on Distributed Computing Systems Workshops (ICDCS Workshops) - Beijing, China (2008.06.17-2008.06.20)] 2008 The 28th International Conference

Improving routing reliability on wireless sensors network withemergency paths

Bastien Mainaud, Mariem Zekri and Hossam AfifiWireless Networks and Multimedia Services Department

of the National Institute of Telecommunications, 9, Charles Fourier - 91011 Evry Paris France{bastien.mainaud,mariem.zekri,hossam.afifi}@it-sudparis.eu

Abstract

One of the most important issues in Wireless SensorNetworks is reliable communication. In this article,we present MAODV-SIM, a routing protocol thatincreases the reliability in a sensor network and whichis based on AODV. Our protocol introduces multipathand signal intensity metric. We need to define whatreliability is? How can we say that a protocol ismore reliable than another? Which benchmark canbe used? Multipath is needed to determine emergencyroutes that are required to find a new path to thedestination if the first one fails. Then,to choose morereliable routes we use a different metric, the signalintensity metric which is based on 802.15.4 standardthat defines two physical measurements : The LinkQuality Indication(LQI) and the Energy Detection(ED). Simulations show that MAODV-SIM increasesreliability in sensor network.

Index Terms - Wireless Sensor Network, RoutingProtocol, Reliabilty, Multipath

1 Introduction

The need for reliable and robust communications isstrong in WSN [1]. In case of medical application, com-munications should not be interrupted or corrupted.Sensor networks are considered to be the future mea-suring devices making the monitoring more comfort-able for the patient,scalable for the medical staff andcost effective [11]. Reliability can be studied on dif-ferent layers of the OSI model, in our case, we chooseto work on the low layers, especially the routing layer.In a classical wired network, the physical environmentdoes not generally interfere with data transmission.This is not true in a WSN where Sensor nodes could

be out of order due to the nature, and a link betweentwo nodes can simply disappear. As the network usesa hop-by-hop communication, node failure may lead toglobal network failure. The need for reliability is hencequite important. The goal of our solution is not todefined an efficient routing protocol, a shortest pathprotocol, neither a low consumption routing protocol,but a reliable routing protocol. We develop a solu-tion using multipath and strength intensity messagemetrics. Moreover, the solution we present shows alsogood performance in the other criteria.

The paper is organized as follows. Section 2 presentsthe notion of reliability and some related works whichdeals with reliable protocols. Section 3 describes oursolution. Section 4 presents some experimental results.Conclusion is presented in section 5.

2 Background

2.1 Reliability

Given the nature of the application, it is absolutelycritical that a query update reach the sensors in a re-liable manner. Any sensor network that is deployedto cater to a critical application, in both civilian andmilitary environments, will require mechanisms to en-sure reliable delivery of information from the sink tothe sensors [12].First, we need to differentiate the reliability of a WSNnode and the reliability of the entire WSN. The firstcase is the capacity of a node to be operational as longas possible. The constraints are multiple. First thedownloaded software on the node must be safe andclean of error (debugging is not possible on a patient).Second, the power consumption is also strongly im-portant : a code optimisation effort must be done forreducing the power consumption and so increasing thenode lifetime. A node which is off is a node that cannot

The 28th International Conference on Distributed Computing Systems Workshops

1545-0678/08 $25.00 © 2008 IEEEDOI 10.1109/ICDCS.Workshops.2008.66

545

Page 2: [IEEE 2008 28th International Conference on Distributed Computing Systems Workshops (ICDCS Workshops) - Beijing, China (2008.06.17-2008.06.20)] 2008 The 28th International Conference

be considered as reliable.The second case is the capacity of an entire WSN torealise the task it was designed for as long as possi-ble. The IEEE defines it as “the ability of a systemor component to perform its required functions understated conditions for a specified period of time “[6].The power consumption topic is, as for the first case,strongly important. If all nodes increase their lifetime,the entire network will increase its own lifetime. Inthis article, we study the entire network lifetime case.Since, we must quantify the reliability, we try to de-rive an objective paramater for it. How can we saythat a protocol is more reliable than an other. What isthe benchmark we will take into consideration for thispurpose. In this paper we essentially use the packetdelivery ratio criterion in order to compare some rout-ing protocols. We choose to base our work on networklayer and to improve the routing protocol in order tobe more reliable.

2.2 Related Work

Multipath routing and algorithms based on energymetrics and their applications have been well studiedin the ad hoc networking literature. In a broad sense,multipath routing enables fault tolerance and metricsbased on signal strength metric increase the networklongevity and improve the route selection decision.An early work [7] by Mahesh K.Marina and SamirR.Das on an application of multipath routing knownas AOMDV discuses how to compute multiple loop freeand link disjoint paths. AOMDV starts from AODVand maintains its principle and its metric. The primarygoal of AOMDV is to provide efficient fault tolerancein the sense of faster and efficient recovery from routefailure in dynamic networks.Another well known example of multipath routing al-gorithm is MOR [2]. Chen has designed the MultipathOn-demand Routing (MOR), an efficient reactive pro-tocol for routing and reliable data delivery in a wirelessad-hoc network. MOR in some ways resembles otherwireless ad-hoc routing and data delivery protocols,but has special characteristics that make it more energyefficient. With MOR, each node forwarding data addsa route back to the sender, and thus by the time thedata has reached its destination, the destination has aroute that can be used for replies or acknowledgments.The most important feature in MOR is that it inte-grates a reliability layer which implements link-layerreliability. The goal of this reliability is to improve thechances of end-to-end packet delivery. There has alsobeen interest in the ad hoc networking community toemploy routing algorithm based on signal strength and

energy.SSR [9] is an on-demand protocol which selects routesbased on the signal strength between nodes and nodeslocation stability. The stronger the signal, the more itindicates that the nodes location is stable. SSR can bedivided into two cooperative protocols: the DynamicRouting Protocol (DRP) and the Static Routing Proto-col (SRP). The DRP is responsible for the maintenanceof the Signal Stability Table (SST) and Routing Table(RT). The SST records the signal strength of neigh-bouring nodes, which is obtained by periodic beaconsfrom the link layer of each neighbouring node. Afterupdating all appropriate table entries, the DRP passesa received packet to the SRP which processes packetsby passing the packet up the stack if it is the intendedreceiver or looking up for the destination in the RT andthen forwarding the packet if it is not.Reliable Energy Aware Routing (REAR) [4] by Has-sanein and Luo is another on-demand routing protocolthat introduces local node selection, path reservationand path broadcasting delay. This solution uses alsoenergy reservation, which means that each intermedi-ate node on the path to the destination will reserve anenergy partition for sending the packet.Signal Stability based on Adaptive Routing (SSA) [3]is another ad hoc routing algorithm which proposes theusage of the received signal strength of packets. It usesthe signal strength for link stability forecasts, assum-ing that links with high signal strength are less likelyto break.

3 On the interest of the multipath

When we speak about multipath, we do not talkabout sending the same data on different paths simul-taneously but to send the data on an emergency routeif the first one fails. There are many reasons becauseof which a packet may not reach its destination. Inthis case, we consider only node failure. When a nodeis sending a packet to another node, the packet willcross a part of the network until reaches its destina-tion. Throughout its journey, it will travel from nodeto node. We suppose that each node on the way has acertain probability of being out of order, so the packetcan not be delivered to its destination.

Now, if we have some emergency routes for the samedestination, we can use these routes to send the packetif the first one is unavailable. In this analysis, we haveconsidered that the path for the same destination coulduse a part of another path (i.e., the paths could not becompletely disjoint).

We used matlab in order to simulate a more exactbehavior of the network using multipath. We create a

546

Page 3: [IEEE 2008 28th International Conference on Distributed Computing Systems Workshops (ICDCS Workshops) - Beijing, China (2008.06.17-2008.06.20)] 2008 The 28th International Conference

Fig.1 - packet delivery ratio under multipath(α=10%)

vector composed of n coefficient where each coefficientrepresents the state of one of the n node. The vectoris filled using α probability in order to know if thenode is out of order. We simulate a 100 nodes sensornetwork and obtain the results gives in figure 1 and2. We used a static topology, the average number ofneighbours for each node is 3. We simulate for differentvalue of node failure : 10% and 50%. Interferencesin the transmissions could also be occurs, but in ouranalysis, we just want to look the behavior of ournetwork regarding node failure.ω represents the packetdelivery ratio, γ represents the number of node in thepath from source to destination and µ is the numberof possible routes to destination.

For 10% of node failure, we can see that most of 75%packets reach their destination with 2 available routes.With 3 routes and more, the packet delivery ratio riseto 90%.

For 50% node failure (Figure 2), the packet deliveryratio is fairly poor. With 50% node failure, if we haveone route for each destination, only 25% of the packetsreach their destination. With 5 routes available, thisratio rises to 50%.

In case of weak wireless sensor network, where nodescan be easily out of order, we see that multipath so-lutions could be a very practical technique to increasethe reliability of packet delivery.

4 Signal Intensity Multipath Routing

In order to improve reliability in a sensor networkand to justify our choice, we choose to start with a rout-ing protocol and propose enhancements over it. Wehave two main contributions in the routing the pro-

Fig.2 - packet delivery ratio under multipath(α=50%)

cess: multipath and energy metric. First we introduceAODV. Our contributions (MAODV and MAODV-SIM) are described afterwards.

4.1 AODV

AODV is a well-know on-demand routing protocol(whose present version is called DYMO [5]), whichmeans that a node will send a demand for a route toits neighbors only when it wants to send a packet. Itwill send a packet named Route Request (RREQ) toits neighborhood. The RREQ packet will be forwardedusing broadcast until the destination, or a node whichknows a route to the destination, is reached. Then aRoute Reply (RREP) is sent in response to the orig-inating node. If the node received multiple RREP, itwill choose the best route using the minimum-hop met-ric. In our solution, we keep the RREQ and RREPconcept, but we modify the response RREP and thebehavior of the sensor nodes.

4.2 MAODV - Multipath AODV

As mentioned in the Related Work section, multi-path routing is a known technique for improving rout-ing. In our solution, we chose multipath in order tohave some emergency routes in case a primary route isunavailable. In classical AODV, when a node receivesmultiple RREP for a same destination, it will keep onlyone route. Each route discovery is associated with highoverhead and latency. This inefficiency can be avoidedby having multiple redundant paths available. And so,a new route discovery is needed only when all knownpaths to the destination break. In our solution, foreach destination, we maintain different routes in therouting table. Each route is obtained when RREP are

547

Page 4: [IEEE 2008 28th International Conference on Distributed Computing Systems Workshops (ICDCS Workshops) - Beijing, China (2008.06.17-2008.06.20)] 2008 The 28th International Conference

received. For each RREP, we add a route in the routingtable. There is no problem of routing table explosionbecause when the route is no longer active, it will bedeleted from the routing table. If the first route wechoose for sending a packet is no longer available, wecan use the next one without needing another route re-quest. This is the first modification we made to AODV. We accept and maintain multiple next-hop routes asobtained by multiple route advertisements. But thisis not enough to improve the reliability of communica-tions. One problem is that different routes to the samedestination may now have different hop counts. howare we going to choose the route on which we will sendthe data?For that, we choose to change the metric used in orderto find the best route.

4.3 MAODV-SIM - Multipath AODV Sig-nal Intensity Metric

In wired networks, the classical metric used is thenumber of hops between the source and the destina-tion. On a WSN, this is not a very relevant metric,because the energy used for sending a packet dependson the real distance between two nodes and not thenumber of hop between the source and the destination.And the reliability of a path to the destination isnot only dependent on the number of hops betweenthe nodes but also on the link state between thenodes of the path. In the previous section, we showthat multipath reduces the packet loss ratio. Nowwe choose another metric in order to increase packetdelivery ratio.In our solution, we choose to use signal intensity asa metric, not to save energy but in order to detectthe best route to the destination: in our case, themost reliable. This metric is a physical informationwhich can be retrieved from the physical layer ofIEEE 802.15.4 devices. This standard defines twophysical measurements which must be realized by thedevice : LQI for Link Quality Indication and ED forEnergy Detection also known as RSSI Receive SignalStrength Indicator. In this paper, we do not discusswhich of these two measurements should be used forour solution (Some papers deal with this topic with inmore depth [10, 8]).

In this paper, we assume that the higher the sig-nal intensity of a received message is, the more reliablethe link between the two nodes will be. In fact, theLQI and ED values are inversely proportional to thedistance between two nodes. There are some otherenvironmental characteristics which affect the values

but with a clear line of sight distance is the onlyparameter which can modify the LQI and ED values.In case of a highly mobile sensor network, this couldbe very important because when two nodes are faraway from each other, the LQI and ED values are low.If the nodes move away, the communication betweenthese two nodes could be broken because the nodeshave gone too far from each other.

Since we do not want to determine the shortest pathbut the more reliable one, the choice of the best route ismade by the source after it receives some route repliesfrom other nodes. The best route is then the route withthe highest signal intensity metric stored in the routingtable.

4.4 MAODV-SIM Protocol detailled

There are two algorithms needed in order to definethe more reliable route. One is required in the sourcenode [algorithm 1] and another one in each node receiv-ing a route reply. We must add a field in the header ofthe Route Reply in order to store the Signal Intensityof the Received Route (SIRR). The overhead of our so-lution is not different from the classical AODV. Thesignal strength is not a complex value, it could be setwith a few bits (1 octet is quite enough)This value is initialized by the initiator of the RouteReply packet to a fixed value. When a node receives aroute reply for a route it did not request (intermediatenode), it compares the value stored inside the route re-ply (SIRR) to the value of the Signal Intensity of theReceived Message (SIRM). This determines the newvalue of the SIRR [algorithm 1].

Input: SIRM,SIRROutput: Reliable Route

if (Node received a Route Reply) thenif (SIRM ≤ SIRR) then

SIRR = SIRM;endif (I’m the initiator of the Route Request)then

add route inside routing table;else

Forward Route Reply to next node;end

endAlgorithm 1: Route Reply Forwarding

When the Route Reply is received by the source, thenew route is added to the routing table, even if thereis already a route to the same destination. The sourcestores the next-hop and the value of the SIRR 1]. So,

548

Page 5: [IEEE 2008 28th International Conference on Distributed Computing Systems Workshops (ICDCS Workshops) - Beijing, China (2008.06.17-2008.06.20)] 2008 The 28th International Conference

for the same destination, we could have multiple paths.For each one of these paths, we have a different valueof Signal Intensity Route : SIRoute. Then, when thesource sends a message, it will choose the best routewith the highest SIRoute instead of the lower numberof hops.

The source will set the value of SIRR to a lowervalue. If a message must be retransmitted (if the for-mer message was lost, due to route failure for example),we must use another route. So, the best route will bedifferent from the former route.

Fig.3 - Example : best reliable route

In Figure 3, we have a sensor network with node Athat wants to send a message to node G. It will senda RREQ to its neighbors. The arrows represent theRREP. The weights on the path represent the StrengthIntensity of the received message (the strength inten-sity of the message received by node E from nodeG is 0.29). Node A will receive some RREPs ,andthereby different routes, to G. According to the newmetric rules, it has to choose the route with the highestStrength Intensity which is A-B-E-G. In other words,A will choose the route which has the least RSSI value.

5 Simulation Results

We implemented and tested our protocol on the NS-2 simulator. We compare our solution to the classicalAODV algorithm. In this part, we attempt to provethat multipath and Signal Intensity Metric increasenetwork reliability. In order to quantify reliability, wechoose to measure Packet Delivery Ratio. With thisratio, we can see how many packets will be lost with,and without, our solution. The higher this ratio (ie:less packets lost), the more our algorithm is reliable.

A sensor node is a weak device which could be out-of-order for many reasons : no more energy, failure incode execution, physical destruction ... In these cases,the network must not be affected, and data must reachthe recipient. We want to see how the network reactswith our solution in case of node failure.

We shut nodes randomly, to simulate node failure.We then observe the evolution of the packet deliveryratio and the end-to-end delay. The routing proto-cols were tested by changing the rate of node failureto account for network tolerance to node failure. Thealgorithms were tested using 50 nodes. The simulationenvironment for all simulations consisted of a 500m by500m region.

Fig.4 - Packet Delivery Ratio Vs. nbr of node failure

Each node was equipped with an omni-directionalantenna.For each protocol, we investigated two perfor-mance criteria: average end-to-end delay and averagepacket delivery ratio.When comparing the performance of MAODV-SIMwith AODV and MAODV, we noticed that, even withthe sharp decrease of packet delivery ratio after losingup to 20% of nodes, MAODV-SIM remains better thanAODV and MAODV [Fig.4]. This is due to the factthat MAODV-SIM chose better, more reliable routesthan MAODV or AODV because of the new metric.Since we choose a more reliable route, there are fewpackets lost.

Improvement in delay is almost always more than80%. This is because availability of alternate routeson node failures eliminates the route discovery latencythat contributes to delay. This is illustrated in figure5.The performance of MAODV-SIM is also consider-ably better when varying the nodes’ mobility. We usea random waypoint mobility model in order to simu-late the behavior of our network.In fact, while moving,if the received signal intensity strength between twonodes remains higher than the threshold, nodes will beable to continue their transmission on the same link.

549

Page 6: [IEEE 2008 28th International Conference on Distributed Computing Systems Workshops (ICDCS Workshops) - Beijing, China (2008.06.17-2008.06.20)] 2008 The 28th International Conference

Fig.5 - End-to-End delay Vs. nbr of node failure

Thus, we will have fewer broken links and betterdata delivery ratio. Figure 6 illustrates the averagepacket delivery ratio of MAODV-SIM in comparisonto AODV and MAODV while changing the nodes mo-bility. There are a few differences between the threesolutions but ours is slightly more competitive.

Fig.6 - Packet Delivery Ratio Vs. Node speed

6 Conclusion, more Future Work

Reliability is a very strong topic in wireless sensornetworks because of the weak nature of the networkand the fragile integrity of the nodes. In this paper,a routing protocol based on AODV is proposed. Inorder to improve the reliability, we use multipath algo-rithm and a new metric : Signal Intensity Metric. Wesaw that packet delivery ratio and end-to-end delaycan be enhanced by our algorithm. As further work,we will implement this solution on tinyOS. This im-plementation enables us to test our solution in a realenvironment. We will also look into other issues re-lated to on-demand multipath routing; for example,

the availability of multiple paths in relationship withnode density and load balancing with multiple paths.

References

[1] I. Akyildiz, W. Su, Y. Sankarasubramaniam, andE. Cayirci. A survey on sensor networks. Commu-nications Magazine, IEEE, 40(8):102–114, Aug 2002.

[2] E. Biagioni and S. H. Chen. A reliability layer forad-hoc wireless sensor network routing. In HICSS’04: Proceedings of the Proceedings of the 37th An-nual Hawaii International Conference on System Sci-ences (HICSS’04), Washington, DC, USA, 2004. IEEEComputer Society.

[3] R. Dube, C. D. Rais, K.-Y. Wang, and S. K. Tripathi.Signal stability-based adaptive routing (ssa) for ad hocmobile networks. Personal Communications, IEEE[see also IEEE Wireless Communications], 4(1):36–45,1997.

[4] H. Hassanein and J. Luo. Reliable energy aware rout-ing in wireless sensor networks. In IEEE Workshopon Dependability and Security in Sensor Networks andSystems DSSNS’06, 2006.

[5] C. I and P. C. Dynamic manet on-demand (dymo)routing, draft-ietf-manet-dymo-12, expires: August10, 2008, 2008.

[6] IEEE Standards Board. IEEE standard glossary ofsoftware engineering terminology— IEEE std 610.12-1990 (r2002), 2002.

[7] M. K. Marina and S. R. Das. On-demand multipathdistance vector routing in ad hoc networks in proceed-ings of ieee international conference on network pro-tocols (icnp), pages 14–23, 2001.

[8] B. Raman, K. Chebrolu, N. Madabhushi, D. Y.Gokhale, P. K. Valiveti, and D. Jain. Implicationsof link range and (in)stability on sensor network ar-chitecture. In WiNTECH ’06: Proceedings of the 1stinternational workshop on Wireless network testbeds,experimental evaluation & characterization, pages 65–72, New York, NY, USA, 2006. ACM.

[9] E. Royer and C. Toh. A review of current routing pro-tocols for ad-hoc mobile wireless networks, PersonalCommunications, April 1999.

[10] K. Srinivasan and P. Levis. Rssi is under appreciated.In Proceedings of the Third Workshop on EmbeddedNetworked Sensors (EmNets’06), 2006.

[11] S. Varshney, U.; Sneha. Patient monitoring using adhoc wireless networks: reliability and power manage-ment. Communications Magazine, IEEE, 44(4):49–55,April 2006.

[12] A. Woo, T. Tong, and D. Culler. Taming the under-lying challenges of reliable multihop routing in sen-sor networks. In SenSys ’03: Proceedings of the 1stinternational conference on Embedded networked sen-sor systems, pages 14–27, New York, NY, USA, 2003.ACM Press.

550