In contrast with those following the individual method, the protocols using the collective method solve the conflicts by changing all the addresses in one of the merging networks. The mechanism is that one of the merging networks has to tell its nodes to release their addresses in order to get new addresses. This method is divided into two approaches according to the way the
2.4 Conflict Resolution Mechanisms
new addresses are acquired - “casual or uniform” changing as depicted in the Figure 2-4.The first phase in the two approaches is identical: the node detecting the merger with another big network has to send a broadcast message in its network. Upon receipt of this message node activity varies according to the approach:
• Casual changing: this is when all nodes in one of merging networks have to change
their addresses and search for new addresses by themselves. Usually, each node from the small merging network has to initiate a new address assignment process. This may, of course, lead in high signaling cost. Moreover, there is no guarantee that the nodes in this network will find a neighboring node possessing free addresses. High la- tency results, because every node need to search in further hops for free addresses. The Prophet protocol [59] is an example of such a case. In it, all nodes in a small network (i.e. one with a small number of nodes), after the merger with a big network, has to release their addresses and to obtain new ones from the large network. Similar- ly to Prophet, the Quorum protocol in [49] forces the nodes in a merging network with few nodes to release their addresses and to obtain new ones from the large net- work. Another example of casual changing of all addresses in one of the merging networks is presented in [61]. In this protocol the nodes form a tree topology, named the address tree. Basically the root node in this tree is responsible for any merger de- cision. This means that any node from a network receiving from a neighboring node an advertisement (Adv) packet including a different MANET ID has to inform its root node by sending a Detection_merge packet. The root (first detector) in this case should communicate with the other root node which belongs to another merger net- work. After a successful connection, the other root should change the addresses of all nodes in its network according to the available free addresses sent by the first detector root. The precise details are that each node in the network will get a new address from its father node wherein the root is the father for k child nodes and each child is a fa- ther for other k nodes. This mechanism suffers from high latency and signaling cost when the child nodes are further than one hop from their father. Moreover, unreliable multi-hop connection may lead to additional issues such as the detection of new mer- ger among the nodes of same network. Finally, this mechanism depends mainly on periodic update between each pair (father, child), which in turn adds additional sig- naling overhead to the network.
• Uniform changing: in this approach the broadcast message includes information on
new addresses of all nodes in one of the merging networks. This means that each node in one network will know its new address when a border node detecting a mer- ger broadcasts a merger message. The efficient way to achieve such change seamless- ly is by making a uniform change to old addresses, e.g. by modifying (i.e. increasing or decreasing) a part of all node IP addresses in one network, thus “uniformly” chang- ing the addresses. Of course, such a uniform approach helps minimize the number of messages needed to change all IP addresses in a network because a node has no need to search for a new address when it receives the merger message instructing it to change its IP address. In addition, this approach supports other functions in MANETs
2.4 Conflict Resolution Mechanisms
such as the routing function. Basically, a route in a routing table becomes invalid if there is any change in IP addresses representing the route entry. If a discovery process is required for the updating of each route in a routing table, such changes lead to a high signaling cost. The key advantage in the uniform approach is that an address re- configuration does not invalidate any cached routes because new changes of IP ad- dresses in each route are known by each node and can be easily modified. Neverthe- less, finding a good means of changing all addresses uniformly is the main issue in protocols following this approach. For example, the authors in [48] present the CoReS protocol in which uniform changing is done by adding offset to all node ad- dresses in one network (the smaller one). However, the solution shows low efficiency because only the cluster head nodes (CR-nodes) are responsible for detecting and solving the duplication resulting from the merger of the two networks, i.e. if only non-cluster nodes (child nodes) from two different networks are within transmission range of each other, they will not be able to detect the merger. In this case each node assumes abnormal cases and sends Warn message to its default CR-node. In the CoReS protocol, if a CR-node receives a Warn message or detects a merger with an- other network two kinds of messages are to be sent through the combined network. First the CR-nodes of the two networks have to synchronize with each other, by ex- changing the information of their tables and parameters. This step is very important to help define the possible address conflicts and select the proper offset by which the nodes belonging to a conflicting rage should modify their addresses. Then, every CR- node that has to change its address range is responsible for informing its child nodes about the required changes. This process takes a long time and increases the signaling overhead in a network. Moreover, the success of the process requires reliable com- munication among the CR-nodes and, also, between each CR-node and its Child nodes. Finally, CoReS has no limitation on the number of bits for an address because it can expand address size as a network grows. This in turn renders the solution in- compatible with the IP architecture which supports a fixed length of 32 or 128 bits. Table 2-3 shows the main features of conflict resolution mechanisms used in MANETs. In this table we can see that uniform changing of address conflicts is preferable to other mecha- nisms because its average signaling cost and reassignment latency is low. Moreover, it requires no coupling with ongoing routing protocols and it allows a node detecting a merger to assign new addresses to conflicting nodes. Finally, because there is no need for mutual exchanges of signals among conflicting nodes to decide which of them is to be changed, the algorithm of this mechanism is low in complexity. However, most auto-configuration protocols which utilize this mechanism are stateful and belong to the centralized approaches which are unsuitable for many scenarios in dynamic ad hoc networks. Therefore, there is a need to find a mechanism which al- lows distributed protocols that follow this mechanism.