Int. J. Electron. Commun. (AEÜ) 66 (2012) 996– 1005
Contents lists available at SciVerse ScienceDirect
International Journal of Electronics andCommunications (AEÜ)
j our na l ho mepage: www.elsev ier .de /a eue
dynamic cross-layer routing protocol for Mobile Ad hoc Networks
ohammed Hawaa,∗, Sinan Taifourb, Mohammad Qasemc, Waleed Tuffahad
Electrical Engineering Department, The University of Jordan, Amman 11942, JordanDepartment of Electrical Engineering & Information Technology, Technische Universität München, Munich 80333, GermanyDepartment of Computer Science, University of Bonn, Bonn 53117, GermanyBoundless Drop Inc., Amman 11181, Jordan
r t i c l e i n f o
rticle history:eceived 30 January 2012ccepted 11 May 2012
a b s t r a c t
A new cross-layer routing protocol, named Dynamic Packet Guidance (DPG), is introduced for Mobile Adhoc Networks (MANETs). Simulation results show that DPG is quite useful for usage in dense networks ofmobile nodes, with medium-to-high speeds, and low-to-medium load. In these scenarios, DPG provides
eywords:ANET
outingradient
a superior performance compared to several well-known ad hoc routing protocols. The low end-to-enddelay and smaller overhead that DPG achieves in such scenarios positively impacts the scalability ofMANETs and reduces the energy requirements of nodes in such networks. DPG also shows immunity tofailing nodes, as it operates consistently almost independently of failing nodes up to a certain ratio.
TS/CTSPG
. Introduction
Mobile Ad hoc Networks (MANETs) are wireless networks thatperate without infrastructure, nor prior knowledge of the net-ork’s topology [1]. Nodes in such networks are free to move
andomly and organize themselves arbitrarily. MANETs haveeceived considerable research attention over the past years dueo their individual characteristics and ease of deployment. Nowa-ays, MANETs find applications in mesh-based mobile networks,ilitary operations, wearable computing, and home networking.A number of challenges face the designers of routing protocols
or MANETs. Such protocols should guarantee the efficient deliveryf data across ad hoc networks while maintaining a minimum com-unication overhead, high throughput and low end-to-end delay.
he designer is faced with bandwidth constraints of the wirelessinks, fading, interference, packet loss, exhaustible energy supply,imited computing capabilities, and a dynamic (rapidly changing)opology.
In response to the above challenges, several ad hoc routing pro-ocols have been proposed in literature [2–18,24–27], which welassify and summarize in the next section. Each and every onef these protocols presents a tradeoff between the different design
bjectives for MANETs, and adds to our understanding of the naturend characteristics of such a dynamic network.
In this paper, we contribute to this understanding by proposinga new reactive wireless ad hoc routing protocol that exhibits uniquefeatures in practical MANET scenarios; specifically when nodes aredense, the network topology is changing fast, communication isbetween a small number of pairs, while small delay and reducedoverhead in delivering data packets are essential.
We assess the performance of the proposed protocol, which wename the Dynamic Packet Guidance (DPG) protocol, to better under-stand its characteristics and points of strength. We also compareit using simulations to three popular MANET routing protocols:Dynamic Source Routing (DSR) [3,4], Ad hoc On Demand Distance Vec-tor (AODV) [5,6], and also the most recent standardization effort,the Dynamic MANET On-demand (DYMO) [7].
The rest of this paper is organized as follows. Section 2 intro-duces the different classes of ad hoc routing protocols, and presentsthe concepts of Route Discovery and Route Maintenance in the con-text of DSR, AODV and DYMO. Section 3 gives a detailed descriptionof the behavior and characteristics of the proposed DPG proto-col. Section 4 presents the simulation setup and the performanceresults for the DPG protocol together with the analysis and expla-nation of these results. Finally, Section 5 concludes our work andtalks about possible future research.
2. Ad hoc routing protocols
2.1. Background
An ad hoc routing protocol dictates how an ad hoc networkshould be logically structured so that data packets can travel overmultiple hops between source and destination nodes. This is done
y running a distributed set of rules that are identical at all nodes.he routing protocol plays a key role in the design of MANETs sincet controls the tradeoffs between the reliability, fairness, scalability,hroughput and latency achieved by the network.
Ad hoc routing protocols can be classified into one of three cat-gories: flooding protocols, clustering protocols and geographicalrotocols. We briefly discuss these categories below:
Flooding protocols are amongst the first and most enduring setf protocols to be proposed for MANETs. In their simplest form,hese protocols deliver data by requiring the source node to broad-ast its packet to all of its neighbors, each of which relays the packeto their neighbors (again by broadcasting), until the packet arrivest the destination, or the maximum number of hops is reached.
Implementing flooding protocols necessitates very little compu-ational complexity and requires very little memory at the variousodes. In addition, flooding protocols can adapt very quickly tony link unreliability or node movement, which makes them quitettractive. However, flooding protocols suffer from large energyxpenditures as extra copies of the same packet are unnecessar-ly sent to the same node by different neighbors. In addition, manyodes not located on the path between the source and destina-ion transmit and receive unnecessary copies of that packet, thusasting their resources.
To improve on the scalability of flooding protocols, flooding cane limited to only a few control packets, which allow the nodeso know and maintain the topology of the network (i.e., build theirocal routing tables). Afterwards, Data packets can be sent as unicastackets from hop to hop to the destination without the need forroadcasting.
Such improved flooding protocols are very popular in litera-ure and can be subcategorized into proactive, reactive and hybridouting protocols:
A proactive protocol (such as the Destination Sequenced DistanceVector (DSDV) protocol [2] and the Optimized Link State Routing(OLSR) protocol [26,27]) continuously learns the topology of thenetwork by periodically flooding topological information amongthe network nodes. Thus, when there is a need to forward a Datapacket to a destination, the routing information to that desti-nation is up-to-date and available immediately. There are twoproblems in such proactive protocols: (a) if the network topologychanges too frequently, the amount of control packets exchangedto maintain the network topology becomes very high and (b) ifthe number of active communicating nodes is low, informationabout most of the network topology will be needlessly collected.On the other side of the coin, reactive routing protocols donot maintain a consistent and up-to-date routing informationto every node in the network. Instead, they find a route onlywhen needed (i.e., on demand) by flooding the network withRoute Request (RREQ) packets and waiting for Route Reply (RREP)responses. This makes sure that the routing overhead scales auto-matically to only what is needed to react to changes in the routescurrently in use. However, such reactive protocols have to slightlydelay the transmission of the first Data packet in a data streamuntil proper routing information is found. The most familiar reac-tive ad hoc routing protocols are: DSR, AODV and DYMO.Finally, in hybrid routing protocols a mixture of the reactive andproactive features are used to exploit specific advantages. Anexample is the Zone Routing Protocol (ZRP) [18], in which a nodemaintains proactively all routing information in its local neigh-borhood, called the routing zone. However, for all destinationsbeyond the routing zone, routes are acquired on demand.
The second category of ad hoc routing protocols is called clus-ering protocols, in which nodes in the network are grouped intolusters, with a cluster head elected for each single cluster [24,25].
un. (AEÜ) 66 (2012) 996– 1005 997
When a node wants to transmit a Data packet, it first sends thepacket to its own cluster head, who sends it to the other clus-ter head, who finally forwards the packet to the destination node.Because of the way these routing protocols operate, cluster headsin such paradigms are expected to have superior processing powerand higher energy reserves. Another drawback of this class of rout-ing protocols is that they have problems catering for movementsand/or failure of nodes, as new clusters need to be formed and newcluster heads need to be elected. Examples of cluster-based rout-ing algorithms include: Low-Energy Adaptive Clustering Hierarchy(LEACH) [12] and Hybrid Energy-efficient Distributed clusteringprotocol (HEED) [13].
Finally, in geographical routing protocols, nodes are presumedto have perfect knowledge of their geographical location in the net-work. The knowledge of the exact location of each node simplifiesthe process of building network-wide routes, as nodes can relay thepackets to their neighbors geographically closest to the destination.However, having each node know its location comes at a price,such as the cost of a Global Positioning System (GPS) hardwareat each node, or the energy needed to continuously run this GPShardware. Examples of such routing techniques include: Greedygeographic routing [14], Greedy Perimeter Stateless Routing (GPSR)[15], End-to-End routing process (EtE) [16], and Beaconless For-warder Planarization (BFP) [17].
In the following sections, we narrow our focus to reactiveflooding-based routing protocols by discussing the most popularprotocols in this category: DSR, AODV and DYMO.
2.2. Dynamic Source Routing (DSR)
DSR uses explicit source routing, in which each Data packet car-ries in its header the complete, ordered list of nodes through whichthe packet should pass [3,4]. This use of explicit source routingallows the sender to select and control the routes used for its ownpackets, supports the use of multiple routes to any destination (forexample, for load balancing or increased robustness), and allowsa simple guarantee that the routes used are loop-free. However,because of source routing, DSR suffers from the fact that the headersize of Data packets increases with increasing route length.
To obtain the routes to arbitrary destinations in the ad hoc net-work, the DSR protocol invokes the two main mechanisms of RouteDiscovery and Route Maintenance. Route Discovery is activated ondemand, that is to say, it is used only when S attempts to send apacket to D and does not already know a route to D.
To initiate Route Discovery, node S transmits a RREQ as a singlelocal broadcast packet, which is received by node S neighbors. EachRREQ identifies the source and destination of the Route Discovery,and also contains a unique request identifier, determined by theinitiator of the RREQ. Each RREQ also contains a record listing theaddress of each intermediate node through which this particularcopy of the RREQ has been forwarded.
If the node is not the target of the Route Discovery, it appends itsown address to the route record in the RREQ and propagates it bylocally broadcasting the packet, with the same request identifier.An exception to this rule occurs if the node receiving the RREQ hasrecently seen another RREQ message from the same source bearingthe same request identifier, in which case the node discards theRREQ.
When node D receives the RREQ, it returns a RREP to the initia-tor of the Route Discovery, giving a copy of the accumulated routerecord from the RREQ. In returning the RREP to the source, node
D will typically examine its own Route Cache for a route back tothe initiator and, if one is found, will use it as the source routefor delivery of the RREP. Otherwise, it can simply (and option-ally) reverse the route accumulated in the RREQ. When node S
9 Commun. (AEÜ) 66 (2012) 996– 1005
ef
dcMuDRs
2
vfafs
dcp
cttnmottitv
rin
auw
2
AcnDHM
tmip
ait
s
downhill (never uphill) in a valley (see Fig. 2).Since the valley around the destination node is virtual, a dif-
ferent valley can be built for each destination to be contacted, and
98 M. Hawa et al. / Int. J. Electron.
ventually receives the RREP, it caches this route in its Route Cacheor use in sending subsequent Data packets to D.
Route Maintenance is the mechanism by which node S is able toetect, while using a source route to D, if the network topology hashanged such that it can no longer use that route to D. When Routeaintenance indicates a source route is broken, S can attempt to
se any other route it happens to know to D, or it can invoke Routeiscovery again to find a new route for subsequent packets to D.oute Maintenance for this route is used only when S is actuallyending packets to D.
.3. Ad Hoc On-Demand Distance Vector (AODV)
The AODV routing protocol is, as the name suggests, a distanceector algorithm. It creates next hop entries at intermediate nodes,or destination nodes that are in active communication [5,6]. When
node is to send a Data packet, it consults its routing table, andorwards the Data packet to the next hop. The next hop does theame, until the packet reaches its destination.
AODV uses destination sequence numbers to ensure loop free-om at all times, avoiding problems such as the Bellman-Fordounting to infinity issue associated with classical distance vectorrotocols.
When a route to a new destination is needed, the node broad-asts a RREQ to find a route to the destination. AODV requires RREQso be disseminated widely throughout the ad hoc network. Whenhe RREQ reaches either the destination itself, or an intermediateode with a fresh enough route1 to the destination, the route isade available by unicasting a RREP back to the originating node
f the RREQ. Each node receiving the RREQ caches a route back tohe originator of the request, so that the RREP can be unicast fromhe destination along a path to that originator, or likewise from anyntermediate node that is able to satisfy the request. The propaga-ion of RREP and RREQ creates next hop entries for the distanceector tables.
In AODV, nodes monitor the link status of next hops in activeoutes. When a link breaks in an active route (say due to a topolog-cal change), a Route Error (RERR) message is used to notify otherodes of that link loss.
AODV does not support multiple routes to the same destinations strongly as DSR; packets normally end up taking the same routentil broken. This means a new Route Discovery has to be initiatedhen a used route is broken.
.4. Dynamic Mobile On-Demand (DYMO)
DYMO was developed as a somewhat simpler design of theODV reactive routing protocol. The goal was to simplify the proto-ol’s implementation, thus lowering the system requirements forodes executing the protocol [7]. DYMO retains the same Routeiscovery principle used in AODV to construct the routing tables.owever, DYMO provides some new features, such as supportingANET-Internet gateway scenarios and path accumulation.Path accumulation is done during the process of transmitting
he RREQ packets through the network. In DYMO, when an inter-ediate node receives a RREQ, it takes note of previously appended
nformation, deducing routes to all nodes the RREQ previouslyassed through, rather than just the originator of the RREQ.
Just as in AODV, detected link failures are made known to the
d hoc network by sending a RERR message to all nodes in range,nforming them of all routes that now became unavailable. Shouldhis RERR in turn invalidate any routes known to these nodes, they
1 A fresh enough route is a valid route entry for the destination whose associatedequence number is at least as great as that contained in the RREQ.
Fig. 1. Forwarding packets from node S to node D via multiple hops of descendinggradient.
will again inform all their neighbors by multicasting an RERR con-taining the routes concerned, thus effectively flooding informationabout a link breakage through the MANET.
3. The Dynamic Packet Guidance (DPG) protocol
3.1. Basic design
The proposed DPG protocol revolves around the concept of agradient. The gradient is a numeric value that represents the dis-tance of a particular node from the destination node measured inminimum number of hops. Each node has therefore a gradient foreach destination it wants to communicate with.
To help visualize how gradients are used in the DPG protocol,suppose node S needs to deliver a packet to node D in Fig. 1. Letus imagine a gradual valley is built around node D (see Fig. 2). Inthis case all nodes will have a particular height in this valley, rep-resented as levels of shading in Fig. 1. This height of each node isits gradient to node D.
In the DPG protocol, packets sent by node S are guided (for-warded) by intermediate nodes downhill with a continuous descentof the gradient to node D. This process is similar to water going
Fig. 2. A visualization of the hypothetical valley built around node D.
M. Hawa et al. / Int. J. Electron. Commun. (AEÜ) 66 (2012) 996– 1005 999
adient
ts
lntpatotitaNow
ttfiCnstdpta
toDi
Fig. 3. Packet forwarding using gr
hey can all co-exist. The destination of a packet defines what valleyhould be followed.
In the DPG protocol, gradients are combined with a MAC-ayer Request-To-Send/Clear-To-Send (RTS/CTS) handshake betweeneighbors. This handshake is used to forward Data packets throughhe network as follows: Each node with a Data packet to send to aarticular destination locally announces that packet to its immedi-te neighbors by transmitting a short RTS message that indicateshe length of the corresponding Data packet, the final destinationf the packet, along with the node’s own gradient value to that des-ination. All the neighboring nodes hear the RTS message and readts contents. Neighbors that have a higher gradient value to the des-ination (i.e., located farther away from the destination) stay silents they are not in a position to participate in packet forwarding.odes with a smaller gradient to that particular destination, on thether hand, are eligible for sending a CTS reply to indicate theirillingness to forward the packet.
To ensure that only one node forwards the packet, the DPG pro-ocol requires each neighbor to wait a small random time (calledhe jitter time) before sending a CTS message. Upon hearing therst CTS, the node that has the Data packet starts transmitting. TheTS indicates that all eligible neighbors should now stay silent andot participate in the packet delivery process. Only the node whoent the CTS successfully will send an Ack for that Data packet, andhen forward the Data packet by announcing it to its own imme-iate neighbors (of course after adjusting the gradient value in theacket). The process continues until the packet reaches the destina-ion. The flowchart in Fig. 3 summarizes the DPG packet forwardinglgorithm.
The RTS/CTS channel reservation scheme achieves three func-
ions in the DPG protocol. First, all nodes hearing either the RTSr the CTS back down and do not transmit for the length of theata packet. This effectively creates a virtual carrier minimiz-
ng collisions on the wireless channel. Remember also that RTS
s and RTS/CTS handshake in DPG.
packets are shorter than Data packet, which minimizes the prob-ability of collision even further. Secondly, since the transmissionof the first CTS silences all other neighbors, this prevents multiplecopies of the Data packet from being forwarded, which limits theenergy and bandwidth consumed by the network in forwarding thepacket.
Finally, the addition of a jitter time before sending the CTSreply allows for a more fault tolerant and scalable network. This isbecause randomness in the jitter time ensures that multiple neigh-bors of a source node participate in forwarding multiple instancesof Data packets from that node, since each such neighbor is equallylikely to succeed in sending the first CTS, thus silencing the otherneighbors and forwarding the packet itself. This makes sure thatvarious Data packets follow different paths through the network,providing an appreciated load balance to the system. In addition,if one of the neighboring nodes fails, another node will send a CTSthus avoiding breaking the whole source-destination link.
We also point out that this CTS jitter time provides an extraflexibility in the protocol design. Say, for example, that some nodeshave very limited resources in terms of energy. If the jitter time is nolonger random; rather it is set to increase in inverse proportion tothe stored energy at the various nodes, a node with a limited powersupply will delay its CTS response until there are no alternativepaths to forward the packet, after which it replies with its own CTS.This ensures that nodes with more energy reserves forward Datapackets instead of power-starved nodes.
Other, more elaborate, variants for using the CTS jitter time arealso possible. For example, if it is undesirable for packets of a singleflow to go through different routes because this might cause out-of-order delivery of Data packets at the destination (thus affecting the
performance of the applications running on top of the network),then the value of the CTS jitter time can be set to a non-randomvalue that is calculated based on the MAC address of the interme-diate nodes, which will result in predictable and repeatable routes
1 Commun. (AEÜ) 66 (2012) 996– 1005
to
3
tctbn
rAncigt
ttpsitp
RiDgciDphic
3
a
•
ah
000 M. Hawa et al. / Int. J. Electron.
hroughout the ad hoc network, yet will reserve the fault tolerancef the DPG protocol.
.2. Route Discovery and Maintenance
In DPG, nodes in the ad hoc network store their gradient valueso the destinations they wish to communicate with in a gradientache. To populate such a gradient cache, a node S (that wantso send a packet to node D) initiates a Route Discovery processy sending a RREQ packet, which is broadcast across the wholeetwork (just as in AODV).
When the RREQ reaches the destination D, the destinationeplies with a RREP which is also broadcast across the network.s the RREQ and RREP packets traverse the network, intermediateodes note how many hops the packet has traveled, and hence cal-ulate their gradient2 (expressed in minimum hop counts) to thenitiator of the packet (S for the RREQ, and D for the RREP). Thisradient value is saved in their respective gradient caches keyed byhe initiator of the packet for later use.
When the source receives the RREP it has been waiting for, itags the Data packet it wants to send with the gradient value tohe destination, which it newly acquired. It then forwards the Dataacket so it can be guided by the gradient to the destination. For sub-equent packets, the node first checks its gradient cache. If an entrys found for the destination, the packet is tagged with the destina-ion’s gradient value and transmitted, otherwise a Route Discoveryrocess is initiated.
The DPG protocol has the advantage that it does not initiateoute Discovery very often (like in AODV and DSR). This is because
t performs Route Maintenance dynamically every time it forwardsata packets. In other words, Route Maintenance in DPG is pig-ybacked on top of Data packets. This is done by adding a hopount filed in the header of Data packets. This hop count field isncremented, hop by hop, and intermediate nodes forwarding theata packet (and also those hearing the RTS/CTS handshake for thatacket) update their gradients to the sender of the packet. Thisas the effect of keeping the routes alive even when the topology
s changing, assuming nodes are communicating in pairs, and areommunicating often enough.
.3. Final thoughts
Some technical issues arise in the DPG protocol and areddressed as follows:
In a wireless LAN, an Access Point (AP) sends broadcast packetsusing a basic rate (e.g., 2 Mbps in IEEE 802.11b or 24 Mbps in IEEE802.11g) rather than a high transmission rate. This is done so thatbroadcast packets can reach a wider range of devices (at fartheraway distances) for a given signal-to-noise ratio of the channel.We call this scenario (where one AP is supposed to communicatewith all nodes in the network) a global broadcast scenario. Suchglobal broadcast concept is incompatible with MANETs, where itis assumed that the nodes communicate with each other throughmultiple intermediate nodes, because of the wide area of theMANET and the limited power transmission capabilities of thenodes. Hence, when the nodes in DPG send broadcast packets,they send them using the same transmission rate for sending
unicast packets, and hence a broadcast packet travels exactly thesame distance as a unicast packet, and presents the same per-formance in terms of data rate. We call this the local broadcast
2 This is done using a hop count field in the packet’s header, which is incrementedt each forwarding node. Nodes receiving a packet multiple times save the smallestop count they receive as the initiator’s gradient value.
Fig. 4. Illustration of RTS/CTS handshake in DPG.
scenario. The purpose of local broadcast is to deliver a Data packetfrom a node to its immediate neighbor (not the final destination)in one hop.
• Combining local broadcast with the RTS/CTS handshake results inan important advantage for the DPG protocol, which is avoidingblocking the whole network. This is illustrated in Fig. 4, in whichnode A is trying to forward a Data packet to node B. We can seethat nodes C and E hear the RTS from node A but do not send aCTS because they have a higher gradient to D compared to A. Thismeans that nodes F and G are not disturbed by a CTS and are freeto transmit since they will not interfere with A’s transmission.This is actually applicable to all parts of the network betweennode A and the source S, preventing most of the network frombeing blocked by A’s local broadcast.
• It is important to realize in Fig. 4 that even though C and E do nothear the CTS from B, they are not expected to send a CTS as theirhigher gradient will force them to stay silent. Only nodes betweenA and D (i.e., those with smaller gradients) will attempt to send aCTS to A. Since these nodes lie in a narrow region between A andD (located to the left of A), such nodes will hear each other’s CTStransmission, and the chance of A receiving multiple CTS packetsfrom its neighbors is extremely minimized. For example, in Fig. 4,the only nodes that will attempt to send a CTS to A are nodes Band M. In the figure, node M hears B sending a CTS and refrainsfrom sending one. What if a node can hear the RTS from A butnot the CTS from B, and hence decides to send a CTS of its own?This is highly unlikely because of the above mentioned reason,but DPG overcomes this by mandating that once a node hears theData packet from A (because B sent its CTS), that node should gointo silence. Finally, we note that if a collision happens, no Ackwill arrive from B, and A will try again.
• To solve the problem of stale gradient values, the time at which anentry has been added to the gradient cache (EntryTime) is noted.Prior to using any gradient value, EntryTime is compared to thenode’s current time (Time); if a particular timeout has passed(ENTRY TIMEOUT) the entry is deleted.
• When a node sends an RTS request multiple consecutive timeswithout receiving any CTS, this is interpreted as either a case ofrepeated collisions due to congestion in the network (which is arare case due to the small size of the RTS packet), or that the topol-ogy has changed significantly such that no next hop can be foundwith an appropriate gradient value. Hence, when the number of
RTS attempts exceeds a preset value of MAX FAILURE COUNT, thegradient to that destination is deleted from the node’s gradientcash. A new Route Discovery is initiated when packets toward
M. Hawa et al. / Int. J. Electron. Commun. (AEÜ) 66 (2012) 996– 1005 1001
Number of nodes, N 25, 30, 40, 50, 55, 60, 65, 70, 75, 80, 90 80, 21, 2, 3, 4,
24, 28
3
toh
enpia
tahstD
tbOmtIptin
tnwgannC
vttci
ol
i
Average node speed, vmax/2 (m/s) 1, 4, 7, 10, 13, 16, 19Rate (packets/s) 0.3, 0.5, 0.7, 1, 1.5, 2Number of failing nodes, n 0, 2, 4, 8, 12, 16, 20,
that destination are received again. If a next hop is found for aparticular destination, the number of previous failures is deleted.
.4. Comparison with DSR, AODV and DYMO
DPG, AODV and DYMO are similar in that the routing informa-ion is saved at individual nodes, and not provided in the headerf the Data packets. These protocols, therefore, have fixed-lengtheaders.
AODV and DYMO use unicast for both RREP and Data pack-ts (based on the inferred routing table at each intermediateode), whereas DPG uses broadcast for RREP, and unicast for Dataackets.3 This, however, does not mean excessive routing overhead
n DPG since the RREQ and RREP are not sent as often as in AODVnd DYMO.
In addition, to save on overhead, Data packets in the DPG pro-ocol can be transmitted using a broadcast MAC address, not thectual MAC address of the neighboring nodes. This saves the over-ead of resolving logical addresses to physical addresses (using,ay, the Address Resolution Protocol (ARP)). It works because onlyhe node that generated a CTS response reads the correspondingata packet, while all other neighboring nodes stay idle.
AODV and DYMO save a particular next hop for each destina-ion. If this next hop goes down or moves out of range, the route isroken and AODV or DYMO have to restart a new Route Discovery.n the other hand, DPG does not save any next hop, and any ofultiple nodes which have the correct gradient value can serve as
he next hop, thus avoiding the problem of a single point of failure.n addition, the choice of which of these nodes forwards the Dataacket depends on who successfully sends the first CTS, which inurn depends on the random jitter time. This results in load balanc-ng as Data packets may randomly traverse different routes in theetwork.
DPG and DSR are similar in that multiple routes are possible forhe same destination, therefore a broken route does not require aew Route Discovery. Nevertheless, DPG and DSR deal differentlyith a failing node along the path. In the case of DSR, a RERR is
enerated back to the source, and the source itself has to attempt new route. This causes extra delay. In the case of DPG, the failingode does not affect the operation, as nodes are not aware whichode will be their next hop until an RTS is sent and a correspondingTS is received.
One exception is when DSR is implemented with Packet Sal-aging. In this case, the node at which the break occurs tries to fixhe route from its local cache. The node is, however, still requiredo send a RERR to the source, which itself resends the packet. Thisreates unneeded overhead in the network, which does not appearn the case of DPG.
DPG, unlike DSR, has a fixed header size. This makes it capablef running in networks with a higher diameter, whereas DSR isimited to a small diameter of 5–10 nodes. Also, routing information
3 It is considered unicast because each DATA packet is forwarded by only onemmediate neighbor.
is distributed throughout the network in the case of DPG, whereasin the case of DSR only the routing information at the source is used.
4. Performance analysis
To quantify the performance of DPG we compare it to the DSR,AODV and DYMO protocols, using solidly defined metrics. The rea-son for choosing these protocols for comparison is that they have allbeen adopted by the Internet Engineering Task Force (IETF) as stan-dards for MANETs. In addition, a lot of the ad hoc protocols proposedin literature have been compared to one or more of these protocols[19–22], which made these protocols the de facto measuring stickfor ad hoc routing algorithms.
4.1. Simulation parameters and metrics
We use a MANET with N mobile nodes moving in a150 m × 100 m simulation area. The number of nodes will be variedin different simulation runs. This will change the valency of eachnode (i.e., the number of neighbors a particular node is capable ofcommunicating with directly). The transmitted power and receiversensitivity for each node is set such that the maximum distanceover which two nodes can communicate is r = 40 m. The wirelesslink data rate is assumed to be 11 Mbps.
All nodes are assumed to be mobile nodes, and the movementmodel used for the simulation is the Random Waypoint Model[3,21], in which each node starts at a random position within thesimulation area, selected uniformly. Each node then selects a ran-dom destination within the simulation area, selected uniformly,and sets its speed toward that destination to a random speedselected uniformly up to a defined maximum vmax. When a nodereaches its destination, it selects a new destination and a new speed,and repeats the process. Under this model nodes appear to wanderthrough the simulation area.
The transmission rate is defined as the number of Data packetssent from each source per second. The rate expresses the originatingdata traffic in the network. The Data packet size is fixed to hold 750bytes of useful data. We set two nodes to communicate as a pairand these two nodes send packets in both directions at the sametime with the same rate.
To assess immunity to node failures, n nodes out of N are madeto fail at random and uniformly chosen times between zero andsimulation time. A failing node does not become active again duringthat particular run. This introduces interruptions in the establishedend-to-end paths between nodes in the simulation. We call theratio of the number of failing nodes n to the total number of nodesN the ratio of failing nodes.
In our simulation, we vary the parameters in Table 1 (one at atime), while maintaining the other parameters in the table to theirdefault values. This way, we test the effects of these parameters onthe DPG protocol behavior.
The metrics used to compare the various protocols are explained
below. They have been chosen to reflect practical properties of theprotocols being tested.
Data delivery ratio (DDR): The data delivery ratio is the ratiobetween the number of Data packets originated by the source’s
1 Commun. (AEÜ) 66 (2012) 996– 1005
afii
hhtFtp
ioaoi
tpsabioaba
tstosc
bstcaae
4
cs
ltdcDob
spw
iwtfo
20 30 40 50 60 70 80 900
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Number of Nodes (node)
Average Delay (sec)
DSRDYMOAODVDPG
DSR is unique in that its energy consumption increases dra-matically against speed. Looking at the NRL of DSR against speedgives insight on the reason; more speed means a higher rate of
2
4
6
8
10
12
Standard Deviation of Delay (sec)
DSRDYMOAODVDPG
002 M. Hawa et al. / Int. J. Electron.
pplication layer and the number of Data packets received by thenal destination’s application layer [21]. The DDR is important as
t describes the maximum end-to-end throughput of the network.Overhead and normalized routing load (NRL): Several researchers
ave used the normalized routing load to quantify the network over-ead [20,21]. NRL is defined as the number of routing packetsransmitted for every Data packet delivered at the final destination.or routing packets sent over multiple hops, each transmission ofhe packet (that is, each hop) counts as one transmitted routingacket.
The NRL measures the degree to which the network will functionn congested bandwidth environments, and its efficiency in termsf consuming node battery power. In addition, protocols that send
large number of routing packets can also increase the probabilityf packet collisions and hence may delay Data packets in networknterface transmission queues.
NRL only counts the number of packets, and not their sizes orheir transmission time, which is important as well. Hence, weropose an additional metric (called the Overhead) to better under-tand the phenomenon of routing overhead. We define Overheads the ratio of non-useful bytes that are transmitted to all byteseing transmitted. In this context, a byte is considered useful if it
s a Data byte, and it is being delivered to the final destination (aspposed to an intermediate node). Data bytes being transmitted ton intermediate node are thus considered part of the non-usefulytes. Bytes comprising MAC headers, MAC RTS/CTS/ACK packetsnd ARP packets are also counted as non-useful bytes.
Delay: The average delay is the average time elapsed betweenhe creation of a new Data packet at the application layer of theource, to the delivery of said packet to the destination’s applica-ion layer. This time includes the Route Discovery process time, ifne is needed. The average is taken over all Data packets that areuccessfully delivered; Data packets that fail to be delivered are notonsidered.
Consumed energy: The consumed energy of a node is calculatedy a simple model which takes into consideration the energy con-umption during transmission and reception of data, and the actualime the node spends transmitting or receiving data. The node’sonsumption of energy is an important metric in ad hoc networks,s energy is a limited resource [1]. The average consumed energyllows for a general comparison between protocols in regards tonergy efficiency.
.2. Simulation results
A large set of results was obtained in our simulations. For spaceonsiderations, we only present a subset of these results that shedsome insights into the features and behavior of the DPG protocol.
The average delay and standard deviation of delay for the simu-ated protocols are shown versus the number of nodes and versushe ratio of failing nodes in Figs. 5 and 6, respectively. The averageelay of DPG is seen to be less than all other ad hoc protocols. Thisan be attributed to the lower number of Route Discoveries thatPG performs during the simulation, mainly due to its capabilityf performing Route Maintenance while transferring Data packetsetween nodes.
The standard deviation of the delay is also low in DPG due to theame reason (see Fig. 6); a low standard deviation means that mostackets are delayed almost the same amount of time, which is truehen no Route Discovery occurs.
The average delay in DSR can be seen to increase with thencreasing number of nodes. This is mainly due to how DSR deals
ith broken routes; it attempts to reach the destination usinghe cached routes at the source, one by one. If all cached routesail, a new Route Discovery is initiated. With increasing numberf nodes in the simulation, and a fast changing topology, there
Fig. 5. Average delay versus number of nodes.
exist more broken routes to try before converging on a validroute.
Fig. 7 shows both the measured Overhead and normalized rout-ing load as a function of rate and average speed.
The routing overhead of DPG is generally always lower than theother routing protocols. This is due to the fact that DPG dependsmainly on Route Maintenance which is piggybacked on data trans-mission, rather than initiating new Route Discoveries every nowand then. In addition, DPG does not use ARP, which results in moresavings.
We also notice from Fig. 7 that more routing overhead is neededas the speed of nodes increases since this results in many routesbeing broken quickly. In addition, we notice that AODV requireshigher routing overhead at extremely low data rates since routesbreak long before a new Data packet needs to be transmitted.
Fig. 8 shows that the lower overhead of DPG results in a loweraverage energy required by nodes running that protocol comparedto energy consumed by nodes running either AODV, DYMO or DSR.This is because lower overhead means nodes need to transmit lessoften, thus consuming less power.
0 0.2 0.4 0.6 0.80
Ratio of Failing Nodes (nodes/node)
Fig. 6. Standard deviation of delay versus ratio of failing nodes.
M. Hawa et al. / Int. J. Electron. Commun. (AEÜ) 66 (2012) 996– 1005 1003
0 5 10 15 20 25 300.4
0.45
0.5
0.55
0.6
0.65
0.7
0.75
0.8
0.85
0.9
Rate (pkt/s)
Overhead (bits/bits)
DSRDYMOAODVDPG
0 5 10 15 20 250.4
0.45
0.5
0.55
0.6
0.65
0.7
0.75
0.8
0.85
0.9
Average Speed (m/s)
Overhead (bits/bits)
DSRDYMOAODVDPG
0 5 10 15 20 250
5
10
15
20
25
Normalized Routing Load (pkts/pkts)
DSRDYMOAODVDPG
0 5 10 15 20 25 300
5
10
15
20
25
Normalized Routing Load (pkts/pkts)
DSRDYMOAODVDPG
uting
tfc
pt
Rate (pkt/s)
Fig. 7. Overhead and normalized ro
opology change, thus more route breaks, thus more tried routesrom the cache. The increasing NRL explains the increase in energy
onsumption.
For the data delivery ratio (see Figs. 9 and 10), DPG sometimeserforms very closely to AODV; as there are many similarities inhe operation of both. However, as the data rate increases, thus
0 5 10 15 20 2540
50
60
70
80
90
100
Average Speed (m/s)
Average Consumed Energy (Joule)
DSRAODVDYMODPG
Fig. 8. Average consumed energy versus average speed.
Average Speed (m/s)
load versus rate and average speed.
forcing the network into saturation (around the 11 packets/s pointin Fig. 10), the DDR of DPG starts to fall slightly below that of AODV.This is to be expected since the RTS/CTS handshake shares the
time with Data packet transmissions, increasing the load on heav-ily loaded networks. However, DPG still outperforms DSR becauseof its smaller header sizes.
20 30 40 50 60 70 80 900.5
0.55
0.6
0.65
0.7
0.75
0.8
0.85
0.9
0.95
1
Number of Nodes (node)
Data Delivery Ratio (pkt/pkt)
DPGAODVDYMODSR
Fig. 9. Data delivery ratio versus number of nodes.
1004 M. Hawa et al. / Int. J. Electron. Comm
0 5 10 15 20 25 300.4
0.5
0.6
0.7
0.8
0.9
1
Rate (pkt/s)
Data Delivery Ratio (pkt/pkt)
DPGAODVDYMODSR
5
rlhcagsc
crda
RpptDD
pgcnt
R
[
[
[
[
[
[
[
[
[
[
[
[
[
[
[
[
[
[
suing his Master’s degree in Communications Engineeringat the Technische Universität München. He is a member ofthe Elite Network of Bavaria. His main research interestsare: information theory, compressive sensing, and wire-less networking.
Fig. 10. Data delivery ratio versus rate.
. Summary and conclusions
In this paper we introduced DPG, a simple wireless ad hocouting protocol for MANETs. The protocol is based on a cross-ayer design combining the concepts of gradients and RTS/CTSandshake. DPG requires minimal processing and memory storageompared to current ad hoc protocols. It is completely on-demand,nd can handle movement of frequently communicating nodesracefully and without introducing re-discovery delays. It washown that it can provide smaller overhead and end-to-end delayompared to DSR, AODV and DYMO.
The DPG protocol also requires less Route Maintenance for ahanging topology at increasing speeds, which saves energy. It alsouns better with increasing valency, as many nodes with equal gra-ient are present, thus packets maintain a route even when nodesre moving or failing.
Future work may include adapting a more intricate method forREQ and RREP forwarding instead of blindly broadcasting suchackets in DPG, which will avoid the well-known broadcast stormroblem [23]. Different algorithms have been suggested in litera-ure to alleviate broadcast storms, which can be combined withPG for further performance improvement, especially since thePG algorithm works better in low-load scenarios.
Another issue that can be researched further is the gains inerformance obtained by forcing the RREP packets to follow theradient to the initiator resulting from the Route Discovery pro-ess. This may save overhead packets to parts of the network thateed not know about the gradient of the destination in order forhe communication between destination and source to work.
eferences
[1] Corson S, Macker J. RFC2501: mobile ad hoc networking (MANET): routingprotocol performance issues and evaluation considerations. Internet Eng TaskForce 1999.
[2] Perkins C, Bhagwat P. Highly dynamic destination sequenced distance-vectorrouting (DSDV) for mobile computers. ACM SIGCOMM Comput Commun Rev1994;24(4):234–44.
[3] Johnson D. Routing in ad hoc networks of mobile hosts. In: IEEE workshop onmobile computing systems and applications. 1994. p. 158–63.
[4] Johnson D, Hu Y, Maltz D. RFC4728: the dynamic source routing protocol (DSR)for mobile ad hoc networks for IPv4. Internet Eng Task Force 2007.
[5] Perkins C, Royer E. Ad hoc on-demand distance vector routing. In:IEEE workshop on mobile computing systems and applications. 1999.
p. 90–100.
[6] Perkins C, Belding-Royer E, Das S. RFC 3561: ad hoc on-demand distance vector(AODV) routing. Internet Eng Task Force 2003.
[7] Chakeres I, Perkins C. Dynamic MANET on-demand (DYMO) routing. InternetEng Task Force 2010 [draft-ietf-manet-dymo-21].
un. (AEÜ) 66 (2012) 996– 1005
[8] Appavoo P, Khedo K. SENCAST: a scalable protocol for unicasting and multi-casting in a large ad hoc emergency network. Int J Comput Sci Netw Secur2008;8(2).
[9] Kim JH, Lee S. Reliable routing protocol for vehicular ad hoc networks. AEU: IntJ Electron Commun 2011;65(3):268–71.
10] Pandey A, Ahmed N, Kumar N, Gupta P. A hybrid routing scheme for mobilead hoc networks with mobile backbones. In: IEEE international conference onhigh performance computing. 2006. p. 411–23.
11] Lee Y, Riley G. Dynamic NIx-vector routing for mobile ad hoc networks. In: IEEEwireless communications and networking conference. 2005.
12] Heinzelman WB, Chandrakasan AP, Balakrishnan H. An application-specificprotocol architecture for wireless microsensor networks. IEEE Trans WirelessCommun 2002;1(4):660–70.
13] Younis O, Fahmy S. HEED: a hybrid, energy-efficient, distributed clus-tering approach for ad hoc sensor networks. IEEE Trans Mobile Comput2004;3(4):366–79.
14] Stojmenovic I, Olariu S. Geographic and energy-aware routing in sensornetworks. In: Handbook of sensor networks: algorithms and architectures.Hoboken, NJ: John Wiley & Sons, Inc.; 2005. p. 381–416.
15] Karp B, Kung H. GPSR: greedy perimeter stateless routing for wireless networks.In: Annual international conference on mobile computing and networking.2000. p. 243–54.
16] Elhafsi E, Mitton N, Simplot-Ryl D. End-to-end energy efficient geographic pathdiscovery with guaranteed delivery in ad hoc and sensor networks. In: Annualinternational symposium on personal, indoor and mobile radio communica-tions. 2008. p. 1–5.
17] Kalosha H, Nayak A, Rhrup S, Stojmenovic I. Select-and-protest-basedbeaconless georouting with guaranteed delivery in wireless sensor net-works. In: Conference on computer communications (INFOCOM). 2008.p. 346–50.
18] Haas ZJ, Pearlman MR. The performance of query control schemes for the zonerouting protocol. IEEE/ACM Trans Netw (TON) 2001;9:427–38.
19] Gupta A, Sadawarti H, Verma A. Performance analysis of AODV, DSR & TORArouting protocols. IACSIT Int J Eng Technol 2010;2(2).
20] Das S, Perkins C, Royer E. Performance comparison of two on-demand routingprotocols for ad hoc networks. IEEE Pers Commun 2001;8(1).
21] Broch J, Maltz D, Johnson D, Hu Y, Jetcheva J. A performance comparisonof multi-hop wireless ad hoc network routing protocols. In: Proceedings ofthe ACM/IEEE international conference on mobile computing and networking.1998.
22] Dressler F. Self-organization in ad hoc networks: overview and classification.ACM Comput Commun 2008;31(13):3018–29.
23] Tseng Y-C, Ni S-Y, Shih E-Y. Adaptive approaches to relieving broadcaststorms in a wireless multihop ad hoc networks. IEEE Trans Comput 2003;52:545–57.
24] Xiang M, Shi W-r, Jiang C-j, Zhang Y. Energy efficient clustering algorithm formaximizing lifetime of wireless sensor networks. AEU: Int J Electron Commun2010;64(4):289–98.
25] Yua J, Qia Y, Wangb G, Gua X. A cluster-based routing protocol for wireless sen-sor networks with nonuniform node distribution. AEU: Int J Electron Commun2012;66(1):54–61.
26] Clausen T, Jacquet P. RFC3626: optimized link state routing protocol (OLSR).Internet Eng Task Force 2003.
27] Clausen T, Dearlove C, Jacquet P, Herberg U. The optimized link state rout-ing protocol version 2. Internet Eng Task Force 2012 [draft-ietf-manet-olsrv2-14].
Mohammed Hawa graduated from the University ofKansas in 2003 with a Ph.D. degree in Electrical Engineer-ing. He received his M.Sc. degree from University CollegeLondon in 1999 and his B.Sc. degree from the University ofJordan in 1997. Dr. Hawa is the recipient of the FulbrightScholarship (1999) and the Shell Centenary Scholarship(1998). He is a published author and a member of theIEEE. He is currently an Assistant Professor of ElectricalEngineering at the University of Jordan. His main researchinterests include: communication systems, wireless net-working, and Quality-of-Service.
Sinan Taifour received his B.Sc. in Electrical Engineeringfrom the University of Jordan in 2009. He is currently pur-
Comm
Waleed Tuffaha received his B.Sc. in Electrical Engi-neering from the University of Jordan in 2009. He iscurrently working as a software engineer at BoundlessDrop. His research interests are focused on computer
M. Hawa et al. / Int. J. Electron.
Mohammad Qasem graduated from the University of Jor-dan in 2009 with a B.Sc. in Computer Engineering with afocus on communications systems. He is currently doinghis master studies in Computer Science at the University
of Bonn focusing on performance analysis. He is also cur-rently working as a research assistant at AG Martini groupfor Computer Networks. His research interests include:performance analysis, machine learning, and ad hoc net-works.
un. (AEÜ) 66 (2012) 996– 1005 1005
vision, machine learning, and wireless networking.
