Fault Tolerance in MANETs: Flooding Vs Checkpointing-Rollback Methods

A survey paper on fault tolerance in mobile ad-hoc networks 1.Abstract Mobile ad-hoc networks are infrastructure less, self-configuring wireless networks. They provide flexibility and scalability which not many other networks can provide. But like in many other networks mobile ad-hoc networks also sufer from a few problems from time to time which may happen due to link failure, failure of nodes or network. In this paper we see two different methods provided, one a flooding approach and the other check pointing and rollback method which guarantee fault tolerance service in mobile ad-hoc networks.

Introduction

MANET's are a group of autonomous nodes or terminals that communicate with each other. They are connected in such a way that it does not rely on any infrastructure so they are called ad-hoc networks. They maintain connectivity of the network over wireless links. As they are not fixed nodes they move in any direction as per the connection requirements. They make a temporary network and keep changing frequently so their topology is di?¬?cult to predict. They use flooding of data mechanism instead of classical routing.

Various network devices can be added to this, with each user having a unique network address. All of these devices that are node now in the MANET network act as a router and can receive and forward data to other nodes. As we have previously stated that they are free to move in any direction depending on the network connectivity needs so they are dynamic in nature. The data transmission on MANET's generally involves multiple nodes so can say it is multi-hop ad-hoc network. Figure 1. Generalized framework of MANET's There has been many advancements in the field of wireless networks but they face more challenges as well. In MANET's, the nodes consumer power for communicating with other nodes and for transmitting data.

If the nodes collectively decide the shortest path or fewest nodes to communicate for the successful transmission of data they save a lot of power. The topology network in MANET's are is called minimum spanning tree graphs. "" The network is said to be k-vertex connected if on removing at most k-1 vertices or nodes even, the network remains connected."" While choosing the k number of nodes for the graph we choose based on the minimum weight. But sometimes due to many reasons nodes might fail so they cannot be selected based on minimum weight.

In that case we select the nodes based on minimum probability distribution function. The second way of addressing the fault tolerance is by checkpointing and rollback recovery. The authors consider a backbone based mobile ad hoc network which is a type of hierarchical network used for scalability and implementation of e?¬?cient protocols. A BCN ( backbone capable nodes) is used as a virtual backbone by selecting it dynamically. These BCN interconnect neighboring BN (backbone nodes). BCN's have more powerful radios and are functionally more capable than ordinary nodes.

The BCN's and other ordinary nodes a?¬?liate with one BN, forming a cluster and BN being its cluster head. The communication uses this backbone nodes and avoids hoping which in turn results in better network performance. If nodes want to communicate with other nodes in the same BN cluster, then routing is straight forward, but if other nodes are in a remote location then the backbone network is used for communication. This mechanism combined with the normal routing protocols provide a fault tolerant communication in mobile ad-hoc networks.In this paper we will see two approaches of to solving these problems, and analyze them to see the advantages and disadvantages for each.

We will see dynamic method with flooding vs hybrid method with distributed clustering approach. Problem formulation: The approach in this method will be by finding nodes for a spanning tree based on some other factors rather than on minimum weighted nodes as any nodes in the network can fail at any point of time due to various reasons. So, we select nodes based on minimum probability distribution function. The minimum spanning tress thus obtained must also be k-connected which means even if at most k-1 vertices are removed, even then the graph must be connected thus preventing the network from failing. We choose based on pdf rather than only minimum weights. We now see the two algorithms that are used in this problem and how they are modified to for a better result.

Performance Analysis

Dijkstra's and Kruskal's are the important algorithms we will use for forming a graph and creating a minimum spanning tree. They are modified little to obtain better results. Here is how Dijkstra's algorithm works. This algorithm is used to find the shortest path from the initial vertex to every other vertex in a mobile graph. The algorithm assumes all the neighboring vertices from the source vertex have a weight of infinite and its weight as 0. Then through iteration it goes to the next vertex from the source which has the lowest weight. Then, from that node it applies the same approach and covers all other vertices in the graph.

It updates the vertices it has covered and weights from the Current vertex to its neighboring vertex. This whole process is basically determined over all the iterations. Kruskal's algorithm is greedy type algorithm that is used to find out minimum spanning tree for a connected graph. In this algorithm, it already has the weights of all the vertices in a graph and selects one which have minimum weight with its neighbor. Then it repeats this process until all the vertex in the graph are covered. It selects vertex so that it does not form a cycle in the graph.

Failure model Mobile network is constrained because Mobile hosts have limited computational resources in terms of processing and storage capacity. Their bandwidth is also very limited and dynamic. The failure of wireless mobile environment is mainly due to mobile hosts because their batter power is highly limited. Crash recovery model for mobile hosts for both ON and BCN is considered. Suppose a mobile hosts crashes then its transmission and recovery is stopped until it is recovered completely. It is considered that the failure rate of BN is less than failure of ON. In a scenario if BN fails then the BCN converts to a BN to keep the backbone connected. The algorithm that is proposed takes care of failed BN's effectively.

Check pointing The nodes within the a cluster belong to the same BN's and can synchronize their checkpointing procedure. A set of checkpoints, one from each node in the cluster forms a consistent local checkpoint for the cluster. If a cluster has failed then it means that one or more nodes in it has failed. Then, from each cluster a set of local checkpoints form a global checkpoint, but they can't coordinate with each other due to feasibility problems because of dynamic nature of MANETs.

So checkpointing at each cluster is independent of others Message passing among the checkpoints in a cluster may be useless if the sender is involved in zig zag cycles. So to detect the formation of z-cycles involving the receiver should know if the sender is available. The protocol that proposed eliminates any control messages in the checkpoint process and also add the dependency information of MH node to the application message.

Did you like this example?

404

21