Modifications to AODV 

Context in source publication

Context 1

... LM informs the routing protocol as soon as d becomes greater than the 0.1 transmission range. The routing protocol then informs the packet source, which stops sending and initiates a route discovery 3 . In this example, packets in transit have 0.1 seconds to traverse the weak link. If the warning comes too late, the weak link breaks before all packets in transit can be salvaged. On the other hand, the warning should not come too early as we want to use the link as long as possible. In [2], Goff, Abu-Ghazaleh, Kahvecioglu, and Phatak implemented a similar mechanism to predict link breakages. This mechanism also measures the signal strength of received packets. A node informs the packet source when the link to the next hop is close to breaking. The packet source initiates a route request but does not stop sending. If the route discovery is successful, the source switches to the new route. Reactive LM temporarily increases the transmission range of a node to reestablish a broken link. Packets in transit can then traverse the reestablished high power link. A node N tries to set up a high power link if the RTS-CTS handshake (to a neighbor N ) X Y with default transmission power is not successful. N therefore sends RTSs with high X transmission power. N must also switch to high transmission power to send the CTS. Y Otherwise, the CTS would not reach N . The RTS frame must therefore contain the X value of the transmission power P . N and N also send the DATA and the ACK t X Y packets with high power. When Reactive LM at N establishes the temporary high X power link to N , it stimulates the routing protocol to begin a new route discovery. Reactive LM maintains a table to record the default and high power links. A node should be able to change its transmission power quickly, because it should not use high transmission power to communicate with neighbors that are within default transmission range. Furthermore, nodes must not broadcast Route Request messages from AODV with high transmission power since new routes should consist of default power links only. As Reactive LM does not use signal strength to estimate the distance to a neighbor, it also raises the transmission power in case of false link failures. However, it can be combined with Persistent MAC, which helps avoid this effect. The IEEE 802.11 MAC Collision Avoidance (CA) mechanism does not work well if nodes have different transmission ranges [6]. Figure 1 shows an example, in which N increases its transmission range from 250 to 300 meters to reestablish the link to 1 N . N does not know about the DATA transmission from N to N , because it could 2 1 5 4 not receive N ’s CTS. N therefore disturbs the DATA transfer from N to N . Due to 4 1 5 4 this effect, as congestion or the load in the network increases, increasing transmit power can in fact degrade performance. Thus, this method should be incorporated only at light loads. Proactive and Reactive LM inform the routing protocol of either weak or high power links. In this subsection, we explain how our modified version of AODV reacts to these MAC layer notifications. The routing protocol does not necessarily have to distinguish between weak and high power links. In both cases, the objective is to inform the packet source of the link failure, initiate a new route discovery and to salvage the packets in transit. In AODV, a route to a destination N in the routing D table of a node N can be in either of two states: The route can be UP, which means N X X forwards packets to N . If N receives a Route Request (RREQ) for N , it will respond D X D with a Route Reply (RREP) because it knows a route to N . The second state is D DOWN; in this state, N does not have a routing entry for N . If N wants to send X D X packets to N , it will initiate a route discovery. If N receives an RREQ for N , it will D X D broadcast the RREQ. If N receives a packet for N , it will drop the packet and X D respond with a Route Error (RERR). We have added an additional route state to allow nodes to salvage packets in transit. This state has the following characteristics: If N receives a packet for N , it X D will forward the packet. If it receives a RREQ for N , it will not reply with a RREP D but broadcasts the RREQ. If an application at N wants to send packets to N , the X D modified AODV will initiate a route discovery. We call this third route state the Going Down (GDWN) state . Figure 2 gives an example, in which the MAC protocol at N informs the modified AODV that the link to N is getting weak (or has become a 2 3 temporary high power link). The modified AODV then, sends a GDWN packet to its active neighbor N , which also sends a GDWN packet to N . All three nodes N , N , 1 0 0 1 and N change the route state for destination N from UP to GDWN. N and N keep forwarding TCP transit packets towards N , but N stops sending packets and initiates 4 0 a route discovery. The old route to N via N , N , and N is then no longer used and D 1 2 3 finally times out, i.e. the route state is set to DOWN. If the MAC protocol reports a link breakage, the modified AODV behaves like the original AODV, i.e. it brings down the route to destination N , sends RERR messages to its active neighbors, and D drops all packets in transit to N . The methods we presented in the previous subsections are aimed at reducing the number of packet drops. TCP Tahoe, Reno, and New Reno grow the congestion window until packets are dropped. In wireless ad hoc networks, congestion does not lead to buffer overflow as in wired networks, but rather to false link failures, which cause the routing protocol to bring down the route. Therefore, even in static networks, where we would expect stable routes, the excessive growth of the TCP congestion window and false link failures cause repeated route changes. This behavior was shown by Saadawi and Xu [12], who quantified the performance of several versions of TCP, in ad hoc networks. They showed that TCP Tahoe, Reno, and New Reno suffer from the “instability problem” due to the excessive growth of the congestion window. They suggested restricting the maximum window size. They also showed that TCP Vegas does not suffer from this instability problem, because it uses a more conservative mechanism with Round Trip Time (RTT) estimations to control the size of the congestion window. TCP Vegas does not need packet losses to stop the growth of the congestion window. Since our goal is to study the effects of mobility as opposed to congestion, we used TCP Vegas for our simulations with ns-2. [3],[10]. The simulation scenario consists of 50 mobile nodes, which move in a rectangular area of 300 by 1500 meters according to the random waypoint model. The random waypoint model defines the node movement as follows: The start position of each node is assigned randomly. A node remains stationary for a specified period of time called the pause time. After the pause time, the node randomly chooses a location and a speed between zero and some maximum speed and travels to that location in a straight line. Once the node reaches its new location, it rests there for pause time seconds and then randomly chooses a new location and travel speed. With a pause time of zero seconds, all nodes are constantly in motion. Figure 3 shows the simulation scenario for two setups that we consider. We call these setup I and setup II. In setup I, there is one TCP Vegas connection between two static nodes. Setup II has two crossing TCP Vegas connections between four static nodes. The static nodes are placed at the edges of the rectangular area. As mentioned earlier, these scenarios are typical when the network is lightly loaded and are appropriate for studying the effects of mobility. When the network is heavily loaded increase in power or proactively searching for new routes might not be appropriate and might increase congestion levels. We re-iterate that our schemes will have to be supplemented by other schemes that can estimate the congestion levels in the network. The traffic over each TCP connection is a file transfer of infinite length, i.e. a TCP source will send TCP data packets for the entire duration of the simulation. The default transmission range of each node is 250 meters and the interference range is 550 meters. A TCP packet travels an average of about 8 hops to get from a source to the corresponding sink. The pause time is zero seconds and the maximum speed of the mobile nodes is set to 0, 4, 8, 12, 16, and 20 m/s for different simulation runs. The simulation time is 600 seconds. These parameters are typically used and are appropriate to study the effects of our protocols with varying levels of mobility. Proactive LM notifies the routing protocol when the distance estimation to a neighbor for current time + 0.1 seconds becomes greater than the default transmission range. This time seems appropriate since we do not want it to be too long (routes are left unused even when the link is fairly stable). The high power transmission range for Reactive LM is 285 meters. This is a system parameter and should be set depending upon the traffic conditions. We choose a reasonable value in this study (approximately a 10% increase from the original range). Table 1 summarizes the simulation parameters. We used the following metrics to qualify the performance of TCP: • Packet loss: Ratio of the number of dropped TCP packets ...