Increasing Packet Delivery Ratio in DSR by Link Prediction

Increasing Packet Delivery Ratio in DSR by Link Preiction Liang Qin Thomas Kunz Department of Systems an Computer Engineering Carleton University, Ottawa, Ontario, Canaa K1S 5B6 {lqin,tkunz}@sce.carleton.ca Abstract Most existing on-eman mobile a hoc network routing protocols continue using a route until a link breaks. During the route reconstruction, packets can be roppe, which will cause significant throughput egraation. In this paper, we a a link breakage preiction algorithm to the Dynamic Source Routing (DSR) protocol. The mobile noe uses signal power strength from the receive packets to preict the link breakage time, an sens a warning to the source noe of the packet if the link is soon-to-be-broken. The source noe can perform a pro-active route rebuil to avoi isconnection. Experiments emonstrate that aing link breakage preiction to DSR can significantly reuce the total number of roppe ata packets (by at least %). The traeoff is an increase in the number of control messages by at most 33.5%. We also foun that the proactive route maintenance oes not cause significant increase in average packet latency an average route length. Enhance route cache maintenance base on the link status can further reuce the number of roppe packets. 1. Introuction In mobile a hoc networks, wireless meia are of limite an variable range, in istinction to existing wire meia. Each mobile noe moves in an arbitrary manner an routes are subject to frequent breakage. During the perio of route reconstruction, packets can be roppe. The loss of packets will cause significant throughput egraation for both real-time an non-real-time ata. This effect on TCP is escribe in [1]. At the time when this research began, few algorithms an protocols trie to improve performance by using link state information. Associativity Base Routing (ABR) propose by Toh [] favors routes with longer-live links accoring to the associativity of the incient noes. He propose a link state preiction moel base on the knowlege of mobile noes position [3]. Flow Oriente Routing Protocol (FORP) [4] is the protocol publishe then that also uses a mobile noe s position information that is provie by Global Position System (GPS) to preict link state. The estination noe can etermine the route expiration time base on the link preiction in the route. It will inform the source noe when a link is preicte to be broken, then the source can select the most reliable route to sen rest packets. The more recent preemptive routing propose by Goof [5] uses signal power strength similar to our metho. It initiates route iscovery early by etecting that a link is likely to be broken an buils an alternative route before the link failure. The preemptive ratio δ is use for efining a preemptive zone that is ajacent to the signal strength threshol. Because δ is a constant, it implies that for mobile noes with ifferent relative spees, the size of the preemptive zone is the same. Their simulation results also show that the increase in overhea coul be as high as three times the overhea of the original DSR protocol. The link availability preiction requires two noes maintain their movement patterns uring the preiction time. Several papers propose probability moel for the link availability [6][7][8]. One moel propose by McDonal [6] efines the path availability as the probability of link is active at time t+t given that it is active at time t. The limitation is that the link coul be broken uring time (t, t+t ), then a packet will be roppe if there is traffic on this link. Another moel propose by Jiang [7] efines the probability of a link that will be active continuously from current time to the time when link is preicte to be broken. In this moel an artificial variable ε is introuce for calculating the path availability, which epens on noe ensity an raio transmission range, an is obtaine from a series of experiments. This paper concentrates on the link state preiction in a hoc networks to reuce the ata packets that woul have been roppe because of link failure. In most existing protocols, a mobile host will keep using the route until the link is broken. Instea, the preiction algorithm propose in this paper will use the power measurement of receive packets to preict the topological change in Proceeings of the 36th Hawaii International Conference on System Sciences (HICSS 3) -7695-1874-5/3 $17. IEEE

orer to perform a route rebuil prior to the link breakage, to avoi the ata packets being roppe. In more etail, a link failure happens when two mobile noes A an B move out of their raio transmission ranges. Noe B monitors the packets coming from A, preicts the link breakage time of link A-B, an then sens a warning message to the source noe of this active route. The source noe can rebuil a new route before the link breaks. We call this pro-active route maintenance. As our simulation results show, pro-active route maintenance reuces the number of roppe packets because of link failure with little averse effect on other routing protocol evaluation criteria (number of control packets, packet latency, an route optimality). Also, the link status information can be use to avoi using stale routes in a noe s route cache to further reuce the number of roppe packets. We o not propose a new routing protocol; instea, we suggest moifications to one popular on-eman routing protocol, DSR [9]. However the preiction algorithm can also be applie to other mobile a hoc routing protocols. This paper focuses on the link state preiction, its implementation an evaluation in the DSR protocol. Section is a brief introuction of the DSR protocol. Section 3 proposes the link state preiction algorithm base on the receive signal power strength, algorithm parameter settings, the preiction accuracy an its limitations. Section 4 presents the preiction algorithm implementation in DSR. Scenarios of exception are analyze. We compare the simulation results uner ifferent mobility rates of DSR with an without preiction algorithm implemente. This section also shows an enhance route cache maintenance base on link status to increase packet elivery ratio. Section 5 raws our conclusion an proposes irections for future research.. Dynamic source routing (DSR) protocol The a hoc routing protocols may be generally categorize as table-riven an source-initiate on-eman riven. The simulation results reporte in several papers [1][11][1] show that normally oneman routing protocols have higher packet elivery ratio an nee less routing messages than table-riven routing protocols. DSR is one on-eman routing protocol. It is base on the concept of source routing, The sener explicitly lists the route in the packet s heaer, ientifying each forwaring hop by the aress of the next noe to which to transmit the packet on its way to the estination host. The DSR protocol consists of two mechanisms: Route Discovery an Route Maintenance. When a mobile noe wants to sen a packet to some estination, it first checks its route cache to etermine whether it alreay has a route to the estination. If it has one, it will use this route to sen the packet. Otherwise, it will initiate route iscovery by broacasting a route request packet. When receiving a request packet, a noe appens its own aress to the route recor in the route request packet if it i not receive this request message before, an rebroacasts the query to its neighbors. Alternatively, it will sen a reply packet to the source without propagating the query packet further if it can complete the query from its route cache. Furthermore, any noe participating in route iscovery can learn routes from passing packets an gather this routing information into its route cache. Each noe along the route, when transmitting the packet to the next hop, is responsible for etecting if its link to the next hop has broken. When the retransmission an acknowlegement mechanism etects that the link is broken, the etecting noe returns a Route Error message to the source of the packet. Then it will search its route cache to fin if there is an alternative route to the estination of this packet. If there is one, the noe will change the source route in the packet heaer an sens it using this new route (this is calle salvaging a packet). When a Route Error packet is receive or overhear, the link in error is remove from the local route cache an all routes, which contain this hop, must be truncate at that point. The source can then attempt to use any other route to the estination that is alreay in its route cache, or can invoke Route Discovery again to fin a new route. 3. Link preiction algorithm The GPS an signal strength methos both use physically measure parameters to preict the link status. The noe with GPS can know the position of itself irectly. But GPS currently is not a stanar component of mobile evices, an in the metropolitan area an inoor, the signal can be too weak to be receive. The signal strength metho only consumes receiving noe s computing power, an oes not epen on any a-on evice. It will be use in this paper. First, we propose a link preiction algorithm that can preict the link breakage time between two mobile noes on an active route. The concept that uses receive signal strength to preict link failure has been propose in cellular networks to perform hanover [13]. At first we assume that the sener power level is constant. Receive signal power samples are measure from packets receive from the sener. From this information it is possible to compute the rate of change for a particular neighbor s signal power level. Because the signal power threshol for the wireless network interface is fixe, the time when the power level rops below the acceptable value can be compute. Proceeings of the 36th Hawaii International Conference on System Sciences (HICSS 3) -7695-1874-5/3 $17. IEEE

3.1. Link preiction algorithm Vay Va In our preiction algorithm, an optimistic raio transmission moel is use. Assume noe A has a transmission range with raius R. If noe B is locate within this circle, it is assume to correctly receive noe A s transmission. So the link availability of two mobile noes can be simply etermine by the istance between them. noe A noe B Vax Vbx 3.1.1. Raio propagation an mobile noe movement moels. Two Ray Groun Reflection Approximation is use as raio propagation moel [11]. At near istances, Friss free-space attenuation (1/r ) is use, at far istance we employ an approximation to Two Ray Groun (1/r 4 ). The approximation assumes specula reflection off a flat groun plane. Because we are only concerne about the situation that two mobile noes move out of their raio transmission range, only the Two Ray Groun moel is use. The signal power strength at the receiver is: Pt * Gt * Gr * ( ht * hr ) P = (1) 4 P: signal power at the receiver. P t : signal power at the transmitter. G t : gain for a signal to a noe from the transmitter. G r : gain for a signal to a noe from the receiver. ht: height of transmitter antenna. hr: height of receiver antenna. is the istance between transmitter an receiver. We assume that P t is a constant (as in IEEE 8.11 raios). Also, in our wireless a hoc network simulation, an omniirectional antenna is use. Further, we assume the groun is flat, an that hr an ht are constants. So we can simplify equation 1) uner the conitions of a hoc wireless network simulation: Pt P = k () 4 Where: k = Gt * Gr * ( ht * hr ) is a constant. This equation shows that the signal power at the receiver noe has relation (1/ 4 ) with the istance between the sener noe an receiver noe. A ranom waypoint moel [9] is use to simulate the mobile movement of noes. A mobile noe will move at a constant spee ranomly selecte up to a given maximum value to a ranom estination, then stops for a preefine pause time an moves again. Let noe B be the receiver noe, noe A be a sener, an also one of noe B s neighbors. Noe A is moving at spee υ a, noe B is moving at spee υ b. υ a can be ecompose to υ ax υ ay, same for υ b, see Figure 1. Then the spee of an noe B relative to noe A is: υ = υ υ ) + ( υ υ ) (3) ( bx ax by ay Figure 1. Mobile noes moving at ranom spees an irections Accoring to equation (3), we can assume that relative to noe B, noe A is still, noe B is moving an at relative spee an irection υ. 3.1.. Preicting link breakage. The preiction time is efine as the time when two noes are moving out of the raio transmission rage. Because the signal power threshol for a wireless network interface is fixe, the preiction time is a constant if two noes keep their moving spees an irections uring this perio of time. noe A θ noe B t Figure. Relative movements of two mobile noes As illustrate in Figure, at time T 1, noe B receives a signal from noe A, assumes the istance between them is, then: Pt (4) = k 4 If noe B knew its spee an irection, relative to noe A, base on equation (4), it coul calculate the current istance an the preicate link breakage time. However, in our moel, we assume that this relative movement information is not known, so we nee to collect further samples to account for these unknown parameters. At time T, noe B receives the secon signal from noe A, let t =T -T 1, Vby t3 Vb t V Proceeings of the 36th Hawaii International Conference on System Sciences (HICSS 3) -7695-1874-5/3 $17. IEEE

P = k P t ( + ( vt ) vt cosθ ) At time T 3, noe B receives the thir signal from noe A, let t 3 =T 3 -T 1, an t 3 is not necessarily a multiple of t (i.e., the spacing in time is arbitrary). P = k 3 P t ( + ( vt3) vt3 cosθ ) At time T, the noe B will receive a signal which power equivalent to the threshol P s, let t = T - T 1, also assume uring time T 1 to T that noes A an B maintain their spees an irections: (5) (6) Pt (7) Ps = k ( + ( vt) vt cosθ ) From equation (4), we obtain: 4 P 1 = kp t (8) Substituting (8) into equations (5), (6) an (7): (9) P = + vt ) vt cosθ P3 = ( ( 3 3 ( vt) vt + vt ) vt cosθ (1) (11) P s = + cosθ From (9) an (1) we obtain: v = β (1) Where: ( P t + P P3 t3 P3 t3 P P3 t ) is a β = ( t t3 t3t ) P P3 constant. From (9) an (11) we obtain: P t (13) P s = ( t P t P + t ) + ( t P t t P t) v Substituting (1) into (13): P t (14) P s = ( t P P t + t) + ( P t t P t t)β We can reformulate this into the equation: at + bt + c = (15) Where: a = t P P s β 1 P ) t P ) b = P (( P β s c = t P Ps t P Finally the preiction time is: b b ac t = ± 4 (16) a Because t is not negative (we are only intereste in future preictions): b 4ac b t = (17) a In the rest of the paper, we call the calculation of link preiction time base on equation (17), using three receive packets signal power strength, our preiction algorithm. With more sophisticate raios, power management may be applie to both sener an receiver so that the power will be controlle with the istance between them. We can moify the original equations an still obtain link breakage preictions in this case, if the source noe attaches its transmission power value with the packet. 3.1.3. Implementing the preiction algorithm. In our implementation, a noe only monitors unicast packets from its neighbors an it is the next hop of the packets, because in DSR, only active routes are maintaine. Most of these unicast packets are ata packets, the rest are Route Reply an Route Error messages. Also, a noe runs the preiction algorithm only when the istance between two noes is bigger than the crossover point (i.e., the istance at which the physical layer uses Two Ray Groun approximation) to save computing power. From the preiction algorithm, at least three packets are neee to correctly preict future isconnection to the neighboring noe. In the implementation, each mobile noe will keep a table (calle signalinfo Table). Each entry of the table hols information such as signal power strength an reception time, of three packets from the same neighboring noe. When noe B receives packets from noe A, it upates its signalinfo array accoring to: P 3 P P 1 an T 3 >T >T 1 (18) When two mobile noes are moving closer, the latest signal power strength will be greater than the previous one. In this case, we set P 1 to the latest signal power value an set P an P 3 to zero, no preiction is necessary. The preiction algorithm assumes that uring T 1 an preiction time T two noes maintain their spees an irections. In reality, two noes may change their spee or/an their irection uring this time. Accoring to the algorithm, the preiction base on the first one or two packets just after the changes may not be correct an cause a false preiction. Our algorithm also assumes there is no noise uring the packet transmission. This is a simplistic assumption. Some experiments that a ranomly prouce noise (5% to 1% of signal power strength) to the signal have been conucte. The simulation results show that the preiction accuracy, as is to be expecte, suffers. We also conucte some experiments to improve the preiction by applying regression metho on the receive signal power strength. This example shows one of effective ways to improve the accuracy in real application environment. More research an experiments are currently being one on this topic. The purpose of our research was to explore whether pro-active route maintenance has the potential to reuce roppe packets by making using of link status information. So we assume in the following the case of a Proceeings of the 36th Hawaii International Conference on System Sciences (HICSS 3) -7695-1874-5/3 $17. IEEE

noise-free environment, recognizing that this will result in optimistic results. All the moification an simulation we i in this paper is base on The Network Simulator (NS). NS is a iscrete event simulator evelope by the University of California at Berkeley an the VINT project [15]. The Monarch research group at Carnegie-Mellon University evelope support for simulation of multihop wireless networks in NS [1]. It provies tools for generating ata traffic an mobile noe mobility scenario patterns for the simulation. DSR is one of the a hoc routing protocols that have been implemente. In NS, the Distribute Coorination Function (DCF) of IEEE 8.11 for wireless LANs is use as the MAC layer protocol. The raio moel uses characteristics similar to a commercial raio interface, Lucent s WaveLAN [16], which is moele as a share-meia raio with nominal bit rate of Mb/s an a nominal raio range of 5 meters with omniirectional antenna. Because any on-eman a hoc network protocol nees a perio of time to set up a new route, a time parameter critical time t s is introuce. t s shoul be big enough for noe B to sen a route error message to the source of the packet an the source to fin a new route. When noe B receives a unicast packet from noe A at time T r, it calculates the preiction that the link between noe A an noe B will be lost at time T. If T - T r t s, we say the link enters the critical state. We implemente the preiction algorithm in DSR, an ae preiction time to the packet trace format. When two noes, which are on a link of an active route, are moving away, the link breakage time will be printe in the tracefile when the receiver receives a packet. Then we compare the preicte value with the time a packet really roppe because of a broken link. Base on these experiments, we foun that active links (efine as links with packet transmissions uring the last secon), more than 9% of the roppe packets are preicte accurately, no matter what the mobility patterns are, unlike the preemptive ratio in [5], which is erive empirically. Because noes move ranomly, two noes may be moving closer though they were moving apart when the preiction is mae, then a false preiction will be mae. Choosing smaller critical time t s can reuce the possibility of false preictions. Though the most appropriate value will be ifferent for ifferent network configurations, in the remainer of the paper, t s is set to 1 secon. 4. Preiction algorithm application After verifying the accuracy of the preiction algorithm, we implemente the pro-active route maintenance in DSR. In the original DSR protocol, the source noe will keep using one route until the link is broken or Route Reply an Route Error messages are receive. Applying the preiction algorithm to the DSR protocol, if a noe preicts that the link with its neighbor that it receive a packet from will break, it can inform the source noe of the packet. Then the source noe can fin a new route to sen other packets to avoi packets being roppe over a broken link. 4.1. Preiction of link status As escribe in Section 3, every noe maintains a table that contains the information of receive packets (previous hop noe aress, packet signal power an reception time) from its active neighbors. Every time a noe receives a unicast packet, it puts upate packet information in its table an upates the preiction time (obviously, if two noes maintain their movement patterns, the preiction time will be the same). At the beginning, every preiction time for the link with an active neighboring noe is set to infinity, inicating that the link will be vali all the time. The preiction algorithm uses a sliing winow with a winow size of three. Every time noe B receives a packet from noe A, it moves the winow one forwar to compute the preiction time by using the latest three packets with ecreasing signal power strength accoring to conition (18). When noes are moving closer, the receive packet s signal power value will increase. Then the preiction algorithm will set the winow size to one an preiction time to the efault value. When it receives the next packet with a receive signal power value lower than the last packet s, no preiction is possible yet (two packets are not enough for getting an accurate preiction). In this case, the preiction time will also be set to a efault value, but we re-open the winow to three again. If two noes are moving apart but suenly slow own or spee up, the preiction algorithm will yiel inaccurate results until a total of three new packets transmitte after the change are receive. 4.. Initiating route error messages Every mobile noe maintains a table calle Link Table. Each entry of the Link Table has the source noe aress, previous hop noe aress, an the packet reception time. Because ifferent routes may share a common link, there may be entries that have the same previous hop noes but with ifferent source noes in the Link Table. Maintaining the source noes in the Link Table can prevent noe B from repeately sening Route Error messages to the same source noe in case it receives more than one packet from noe A after the link A-B enters critical state. When mobile noe B receives a unicast packet from its previous hop noe A, it will monitors its link status with Proceeings of the 36th Hawaii International Conference on System Sciences (HICSS 3) -7695-1874-5/3 $17. IEEE

noe A using the preiction algorithm. If the link is in the critical state (link breakage preicte in less than t s secons), noe B first will check if this packet is a Route Error message. If it is, noe B will not sen another Route Error message to prevent Route Error messages neig sent back an forth between them (noe A may also have enough information an preict this link will be broken). Next noe B will check if the source noe an previous hop noe pair is in its Link Table. If it is, noe B has alreay sent a Route Error message to the source about the imminent link A-B breakage recently. If it cannot fin an entry, noe B will insert the source noe an previous hop noe pair with the current time into the Link Table. Then noe B sens a Route Error message to the source noe of this packet using the source route in the packet heaer. To ifferentiate from the original Route Error message, we call the latter Preiction Route Error message. As the DSR protocol specifies, when a Route Error message is sent to the source noe, the noes on its route will remove the routes with link A-B from their route caches. 4.3. Hanling route reply messages When a source noe receives the Preiction Route Error message, similar to the original DSR protocol, if it cannot fin an alternative route that oes not contain the flagge link A-B, it will initiate a Route Request message an broacast it. We piggyback the information about link A-B preicte to break soon onto the Route Request to allow other noes in the network to clear their route cache as well. When the source noe sens a Route Request message, the A-B link is still alive, so there are still Route Request messages that are sent over this link. Consequently, some Route Reply messages may have routes that contain the A-B link. When a source noe receives this kin of Route Reply messages, it may use this soon-to-be-broken link again, an a packet still will be roppe. We use the Link Table in every mobile noe to solve this problem. When noe B receives a Route Request message, it first checks if the previous hop noe is in the link table. If it is, it stops propagating this Route Request message further an oes not reply from its route cache. The new routes will not contain the A-B link anymore. Because of the ranom characteristics of the mobile noe mobility, two noes move closer again sometime later. A time-out mechanism is applie to the Link Table entries. Entries will be remove when they time out. As long as the two noes are moving away from each other, one noe cannot be the next hop of the other. If at a later point they are moving closer, they may become neighbors in a new route again. Because we alreay set the critical time to 1 secon, we set the Link Table timeout to secons. 4.4. Experimental results We implemente the proactive route maintenance in DSR an ran simulations. Exploring the results, we gaine aitional insights into the interaction between DSR an pro-active route maintenance. For example, because of the salvage mechanism use in DSR, once the noe salvages the packet, the source route in the packet heaer will be change to the new route foun in its route cache. If a Preiction Route Error message is initiate by one of the noes on the new route, this message will reach the noe that salvages the packet, instea of the source noe. So the source noe is still not aware of the soon-to-be-broken link, if a new route contains this link, a packet will be roppe. A more complicate example is illustrate in Figure 3; the ashe line inicates that there may be more than one hop between two noes. Suppose that noe B alreay sent a Preiction Route Error message to source noe S via noe D, an S receive it. Because noe S ha an alternative route, which oes not contain the ba link A-B, to the estination via noes X an Y, it uses it to sen packets. Sometime later, link X-Y breaks, an noe X happens to have a route in its route cache via Z an containing link A-B. Because noe X is not aware of this ba link. It will use this route to salvage the packet. At this time the link A-B might have broken, then the packet will be roppe. D S Z A Figure 3. Exceptional scenario of pro-active route maintenance in DSR X Timing-relate complications can also occur, to some extent these (or similar) issues exist in the original DSR protocol as well. For example, source noe S receives a Preiction Route Error message, shortly after it receive a elaye Route Reply message which replies to a previous Route Request issue by S. Noe S might use this route to sen packets. If this route contains a ba link that was reporte in the Preiction Route Error message, the packet will be roppe on this link. Base on these an other observations, we set the B Y R Proceeings of the 36th Hawaii International Conference on System Sciences (HICSS 3) -7695-1874-5/3 $17. IEEE

number of Preiction Route Error messages a noe can sen to the same source noe to 3 times, which also increases the probability for the source to receive this message in case of transmission error. Table 1 compares the packet loss rates when sening the Preiction Route Error message once an at most 3 times. The pause time is s, maximum spee is m/s. The entries in the table are explaine below. Table 1 shows that sening multiple Preiction Route Error messages can further reuce roppe packets, but also increases the number of control messages. The total number of packets roppe on the active link is reuce almost by half. We ran simulations with ifferent mobility patterns in NS to verify the pro-active maintenance in DSR. The a hoc networks use in our simulation consist of 5 wireless noes, moving in a rectangular (15m x 3m) flat space. We choose the traffic sources to be constant bit rate (CBR) sources. The transmission rate is 4 packets per secon. Each packet is 64 bytes big. In Tables an 3, the format of scenarios is: number of source noes-pause time-maximum spee (the first row in the table). The Table 1. Comparison of sening preiction route error message times simulation time is 9 secons using NS version.1b6. We mainly choose the high mobility scenarios, because they cause more link breakages. We ae a higher ata traffic scenario with 3 source noes. In total 5 scenarios are simulate, each with seven ranomly generate scenario files. To compare the performance with the original DSR protocol, the following four metrics are use: Total number of ata packets roppe, an the number of ata packets roppe because of no route. Total control message: total number of control packets transmitte. Each hop-wise transmission of a control packet is counte as one transmission. Hop count ratio: The ratio between the total number of hops all packets travel to reach their estinations an the total length of the shortest path that physically existe through the network when the packets are originate. NS will provie the length of the shortest possible path between all noes in the network. Average packet latency: average ata packet latency. Original Once Three Times Percentage % Total Data Packets Droppe 1315 95 87-13.13 Total Data Packets Droppe (No Route) 1153 831 7-15.5 Total Control Messages 673 3465 3371 3.83 Hop Count Ratio 1.88 1.86 1.85 -.1 Active Link (less than 1 secon) 47 75 41-47.89 Table. Simulation results of DSR protocol with preiction algorithm: part 1-3-1-3- -3- Total Data Packets Droppe 658 18 657 Original DSR Protocol 751 371 897 Improvement (%) 3 1.344 4.8 6.979 95% Confience Interval (-1.58, 34.947) (5.73, 4.694) (.776, 33.181) Total Data Packets Droppe (No Route) 9 114 5155 Original DSR Protocol 416 1947 749 Improvement (%) 43.863 37.7 31.8 95% Confience Interval (7.94, 59.786) (7.3, 47.114) (3.17, 38.798) Total Control Messages 1947 57785 1485 Original DSR Protocol 16674 47915 16848 Increase (%) 18.696.44 5.359 95% Confience Interval (3.99, 33.483) (1.68, 9.87) (17.767, 3.951) Hop Count Ratio 1.61 1.77 1.86 Original DSR Protocol 1.6 1.8 1.86 Increase (%) -.159 -.7 -.6 Average Packet Latency Increase (%) 7.48-16.48-16.47 Proceeings of the 36th Hawaii International Conference on System Sciences (HICSS 3) -7695-1874-5/3 $17. IEEE

From Table an Table 3, we raw the following conclusions: Every scenario shows an improvement for the total number of no route roppe packets. The improvement is at least %, an reaches up to 44.95%. These improvements are statistically significant at a 95% confience level. Similarly, the control messages increase up to 33.5%. The improvement ratio for the total number of roppe packets is less than the ratio for the no route roppe packets. There are three cases in the low mobility scenarios (two in -3-1, the other in - 3-) where the total number of roppe packets increase because of packets roppe in the network interface queue of a few noes. The two exceptions also result in a negative overall average for the - 3-1 scenario. However, as shown by the 95% confience interval, this result is not statistically significant. The variation of hop count ratio is below 1%. Similarly, the average packet latency oes never increase significantly, but is reuce substantially in a number of scenarios, espite the increase overall traffic in the network. For the same mobility scenario ( secon pause time an up to m/s spee), a scenario with more sources benefits more from pro-active route maintenance than one with fewer sources. Scenario 3--, compare to --, is one such example. For 3 sources (i.e., 3 active flows), noes have more recent information about links to make goo preictions. Accoring to the simulation results, the preiction algorithm implemente in the DSR routing protocol successfully reuces no route roppe packets by at least %, with a cost of increasing route message by at most 33.5%. For the scenarios stuie, this increase in messages (an therefore network traffic) oes not cause aitional packet losses or packet latency increases. 4.5. Enhance route cache maintenance base on link status In DSR, the route cache plays an important role. If there is a route to the request estination in the route cache, a source noe can use it to sen packets without starting Route Discovery; intermeiates noes can use it to reply to Route Request messages, which will reuce latency an overhea. Also, a mobile noe can salvage a packet by using its route cache if the next hop in the source route is broken. On the other han, because of mobility, the routes in the route cache may be stale. If the mobiles use these routes, packets may be roppe on the broken links. DSR oes not have an explicit mechanism to remove stale routes from the route cache. To evaluate the effect of stale routes on the packet elivery ratio, we analyze the tracefiles of NS simulation. We o not count the percentage of stale routes in each route cache, because there is no guarantee that a noe actually uses a stale route from the route cache. We explore how many stale routes actually are use an also cause packet losses. Figure 4 shows the roppe packet ratio cause by stale routes. Packet loss occurs when the source noe uses the stale route to sen packets, or intermeiate noes use stale route to sen Route Replies an/or salvage a packet along a stale route. We ran simulations for ifferent pause times with maximum spee of 1m/s an m/s respectively, for each scenario we ran two simulations, an then compare the results with the total number of roppe packets because of no route. The result shows that for high mobility, packet loss ue to stale routes accounts for between 3% to 4% of all packet losses; for low mobility, links will be more stable, the ratio becomes lower. We can use the link status base on the preiction algorithm to avoi using these stale routes in the route cache. We have moifie the route cache so that a noe only uses a vali route. In the route cache, every route associates with a route expiration time. When the estination noe sens a Route Reply message to the source noe, it will set the route expiration time (RET) to a big value. When an intermeiate noe receives this message, it also probably can get the link breakage preiction time base on the packets it receive from this sener, if they are moving apart an the preiction algorithm has enough packets to make a preiction. The noe compares the value with RET in the Route Reply message, if this value is smaller, the noe will replace RET with it, an this link is the weakest link in the route so far. Finally the source noe gets a route with RET characterizing the shortest-live link on that route. When a source noe sens packets, or intermeiate noes reply to a Route Request message, they will not use stale routes by checking RET. In our implementation, we only use routes that will be vali for at least 3 more secons. If a noe replies to a Route Request from its route cache, it will sen the route s RET with the Route Reply message. Figure 5 shows our simulation results. Three mobility scenarios are simulate. The format shown in Figure 5 is pause time-max spee. We ran two simulations for each scenario, an the total number of roppe packets is the sum of both runs. For all three scenarios, selecting vali routes from the route cache can reuce by on average 8% the total number of packets being roppe, compare to only using the pro-active route maintenance. Proceeings of the 36th Hawaii International Conference on System Sciences (HICSS 3) -7695-1874-5/3 $17. IEEE

Table 3. Simulation results of DSR protocol with preiction algorithm: part -- 3-- Total Data Packets Droppe 5577 9494 Original DSR Protocol 7481 14476 Improvement (%) 4.67 34.95 95% Confience Interval (17.5, 31.84) (3.17, 38.37) Total Data Packets Droppe (No Route) 468 87 Original DSR Protocol 658 1934 Improvement (%) 7.85 37.466 95% Confience Interval (.63, 35.546) (33.354, 41.579) Total Control Messages 181547 7665 Original DSR Protocol 146714 8191 Increase (%) 3.96 1.7 95% Confience Interval (17.187, 3.664) (17.1, 6.31) Hop Count Ratio 1.79 1.8 Original DSR Protocol 1.81 1.76 Increase (%) -.199.37 Average Packet Latency Increase (%) -5.8 -.53 Droppe Ratio % 5 4 3 1 max m/s max 1m/s 3 9 3 Pause Time (s) strength. It preicts the link status after receiving three packets. When the link enters critical state, the noe will sen a Preiction Route Error message to the source noe of the packet. After receiving a Route Error message, the source noe will initiate a Route Request message an broacast it if it cannot fin an alternative route to the estination an also has packets to eliver. When a noe receives a Route Request, it will check if it comes from an upstream noe of a soon-to-be-broken link. If it is, the Route Request is iscare. Figure 4. Droppe packet ratio because of stale routes Figure 5 shows our first stage of implementing enhance route cache maintenance base on link status information. Further work will be on (1) upating RET values base on new link status, () removing stale routes from the route cache, (3) using probe message to valiate routes before they expire. We expect further improvement to be achieve. 5. Conclusions an future work In most mobile a hoc network routing protocols, sources will use a route until a link is broken or a shorter route is foun. A link failure will cause packets to be roppe an results in transmission elays while a new route is foun. In this paper, a link status preiction algorithm is evelope an ae to DSR. Every mobile noe monitors the receive unicast packets signal power Number of Packets Droppe original with preiction vali cache 5 15 1 5-3- 9- Mobility Pattern Figure 5. The total number of roppe of packets because of no route for original DSR (original), DSR with preiction algorithm (with preiction) an DSR with preiction algorithm an selecting vali route in the route cache Simulations have been one for low an high mobility scenarios. The results show that the total no route packets roppe are reuce by at least %, an up to 44.95%. Proceeings of the 36th Hawaii International Conference on System Sciences (HICSS 3) -7695-1874-5/3 $17. IEEE

At the same time, the total number of control messages increase up to 33.5%, an the hop count ratio staye at the same level. The increase control messages o not cause network congestion, significantly increase average packet latency an packets loss. So pro-active route maintenance can significantly improve overall routing performance. We also enhance the route cache management base on the link status information. The first stage implementation further reuces by on average 8% the total number of roppe packets because of no route. Because the preiction algorithm is general, it can be applie to other mobile a hoc network routing protocols. We are currently implementing the pro-active route maintenance in the A Hoc On-eman Distance Vector (AODV) protocol [17]. We are exploring how to eal with noise. One avenue is to use curve-fitting techniques, base on a winow of recent (noisy) packet reception power levels. We can then base the link preiction on the fitte curve, assuming that the fit is reasonably goo (as expresse by the SSR value). We are also working on (1) soft-hanover: when a source noe receives a Preiction Route Error message, it will continue use the existing route until a new route is foun to further reuce packet latency, () using the preiction algorithm to remove stale routes in route caches by piggy-backing preicte route error message with control massages even if the to-bebroken link is not on the active route, (3) more simulations an analysis on the value of critical time to increase preiction accuracy, an (4) how the reuction of packet loss impacts the performance of TCP in a hoc environments. 6. References [1] G. Hollan an N. Vaiya, Analysis of TCP Performance over Mobile A Hoc Networks, Proc. of the Fifth Annual ACM/IEEE International Conference on Mobile Computing an Networking, Seattle, Washington, August 1999, pp. 19-3. [] C.-K Toh. Associativity-Base Routing for A-Hoc Networks, Wireless Personal Communications Journal, Special Issue on Mobile Networking an Computing Systems, vol. 4, no., March 1997, pp. 13-139. [3] H. Dajing, J. Shengming an R. Jianqiang, A Link Availability Preiction Moel for Wireless A Hoc Networks, Proc. of the International Workshop on Wireless Networks an Mobile Computing, Taipei, Taiwan, April, D7-D11. [4] W. Su an M. Gerla, IPV6 Flow Hanoff in A-Hoc Wireless Networks Using Mobility Preiction, Proc. of IEEE Global Communications Conference, Rio e Janeiro, Brazil, December 1999, pp. 71-75. Conference on Mobile Computing an Networking, Rome, Italy, July 1, pp. 43-5. [6] A. B. McDonal an T. Znati, A Path Availability Moel for Wireless A-Hoc Networks, Proc. of IEEE Wireless Communications an Networking Conference, New Orleans, LA, September 1999, pp. 35-4. [7] S.M. Jiang, D.J. He an J.Q. Rao, A Preiction-Base Link Availability Estimation for Mobile A-Hoc Networks, Proc. of IEEE Infocom, Anchorage, AK, April 1, pp. 1745-175. [8] A. B. McDonal an T. F. Znabi, A Mobility-Base Framework for Aaptive Clustering in Wireless A hoc Networks, IEEE Selecte Areas in Communications (JSAC), vol. 17, no. 8, August 1999, pp.1466-1487. [9] D. B. Johnson an D. A. Maltz, Dynamic Source Routing in A Hoc Wireless Networks, in Mobile Computing, eite by Tomas Imielinski an Hank Korth, Kluwer Acaemic Publishers, Chapter 5, pp.153-181, ISBN: 79396979, 1996. [1] J. Broch, D. A. Maltz, D. B. Johnson, Y.-C. Hu an J. Jetcheva, A Performance Comparison of Multi-Hop Wireless A Hoc Network Routing Protocols, Proc. of the Fourth Annual ACM/IEEE International Conference on Mobile Computing an Networking, Dallas, TX, Oct. 1998, pp. 85-97. [11] S. R. Das, C. E. Perkins an E. M. Royer, Performance Comparison of Two On-eman Routing Protocols for A Hoc Networks, Proc. of the IEEE Conference on Computer Communications, Tel Aviv, Israel, March, pp. 3-1. [1] E. M. Royer an C.-K. Toh, A Review of Current Routing Protocols for A-Hoc Mobile Wireless Networks, IEEE Personal Communications Magazine, April 1999, pp. 46-55. [13] B. Narenran, P. Agrawal an D. K. Anvekar, Minimizing cellular hanover failures without channel utilization loss, Proc. of IEEE Global Communications Conference, San Francisco, CA, December 1994, pp. 1679-1685. [14] T. S. Rappaport. Wireless Communications: Principle an Practice. Prentice Hall, New Jersey, 1996. [15] K. Fall an K. Varahan, eitors. NS Notes an Documentation. The VINT Project, UC Berkeley, LBL, USC/ISI, an Xerox PARC, November 1997, see http://www.isi.eu/nsnam/ns/. [16] B. Tuch, Development of WaveLAN, an ISM Ban Wireless LAN, AT&T Technical Journal, vol. 7, no. 4, July/August 1993, pp. 7-33. [17] C. E. Perkins an E. M. Royer, A-hoc On-Deman Distance Vector Routing, Proc. of the n IEEE Workshop on Mobile Computing Systems an Applications, New Orleans, LA, February 1999, pp. 9-1. [5] T. Goff an N. B. Abu-Ghazaleh etc., Preemptive Routing in A Hoc Networks, Proc. of the Seventh Annual International Proceeings of the 36th Hawaii International Conference on System Sciences (HICSS 3) -7695-1874-5/3 $17. IEEE