New approach for selecting multi-point relays in the optimized link state routing protocol using self organizing map artificial neural network: OLSR-SOM

ABSTRACT


INTRODUCTION
The selection of multi-point relays (MPRs) nodes according to the algorithm described in requests for comments3626 (RFC3626) document [1] is limited to symmetrical connections between nodes. This issue generates several drawbacks in practice on the ground, especially with other parameters that must be integrated into the calculation of these MPRs, namely the nodes energy, the nodes mobility, the nodes geographical position and the bandwidth link between nodes.
In the literature, many researches have tried to develop the selection of MPRs in the optimized link state routing (OLSR) protocol by proposing new approaches [2]. In order to improve both quality and security in selection of MPRs the researchers used several techniques. There are those who treated just the energetic side of the nodes [3]- [7] others treated the effect of mobility [8]- [12]. To integrate more parameters both in the calculation of MPRs and in order to improve the functioning of mobile ad hoc network (MANET) networks precisely of the OLSR protocol, the researchers used techniques (genetic algorithms and neural networks) making it possible to process and combine several criteria at the same time, but, they still remain less efficient because they only integrate some parameters and not all.
The important role that neural networks play in processing and analyzing data for the purpose of classification or prediction has prompted us to conduct a study on a set of approaches [13] which summarizes the use of neural networks in MANETs. Some researchers have focused their study on the use of artificial ISSN: 2252-8938  New approach for selecting multi-point relays in the optimized link state routing protocol … (Omar Barki) 649 neural network (ANN) for the detection of anomalies and attacks in the network [14], [15] and some for the mobility prediction [16]- [18] and the others for optimization of the route from source to destination and traffic estimation [19]- [21]. This study has shown that the different ANN offer a great capacity for modeling [22], [23], classification [24], [25] and prediction [26] of future values on the basis of data collected over time for a given event. In the next section, we will describe the principle of our new approach that uses self organizing map (SOM) ANN classification technique to improve both quality and security in MANET. Finally, in section 3, we will present the simulations that we carried out and then we will present their results and discution.

NEW APPROACH FOR SELECTING MPRS: OLSR-SOM 2.1. Our approach description
Our approach is based on the artificial neural network of SOM classification which allows sorting the MPRs nodes into distinct groups based on data extracted as the simulation progresses. The groups formed are subjected to a filter of criteria that we have defined to determine the strong and weak classes. The MPRs that is belonging to the weak class will be eliminated from the final list of MPRs used later in routing by OLSR protocol under NS3, and so on until the end of the simulation.
This new OLSR-SOM approach is based on the following steps:

OLSR-SOM scheme
The improvement made on OLSR to obtain OLSR-SOM focuses on filtering and reducing the list of MPRs already selected by standard OLSR algorithm presented in Figure 1(a) to a new list of MPRs considered strong to ensure the transmission of the messages and data with high reliablity as shown in Appendix. The changes made to OLSR to get OLSR-SOM are shown the Figure 1(b) as shown in appendix. In Figure 1(b), we have added classification processing on the list of MPRs (already selected by the standard MPRs selection algorithm described in RFC3626) using the SOM ANN in Scilab software for the classification followed by the selection of a new list of MPRs for the node X.

RESULTS AND DISCUSSION
In order to prove the effectiveness of the new OLSR-SOM MPRs selection approach in terms of quality and security, we have run several simulations using NS3 which is an open source discrete event simulation tool. To support our OLSR-SOM MPRs selection approach, we modified the MPRs selection algorithm included in NS-3. These simulations have been carried out to study two cases. The first one is related to evaluate the perfomace measures of the MANET network using this new OLSR-SOM MPRs selection approach and the second one is concerning the performance evaluation of the MANET network against the distributed denial of service attack (DDOS).

Perfomace evaluation of the MANET network using a new OLSR-SOM approach
In order to evaluate the performance of the new OLSR-SOM MPRs selection approach. We consider the following scenario in which the number of network nodes is fixed whereas the number of nodes implemented the new OLSR-SOM algrorithm is increasing. The simulation parameters are given in Table 1.
We run several simulations in order to evaluate some measures of the performance of the MANET network; namely the packet delivery rate (PDR), the throughput, the packet loss and the energy consumption, by increasing the number of OLSR-SOM nodes from 1 to 5 in the network of 10 nodes as indicated in the Table 1. We plot the simulations results in Figures 2-5 representing the evolution over time of the PDR, the troughput, the packet loss and the energy consumption respectively. According to the Figure 2, we note that the PDR parameter at the beginning of the simulation has the same value for the different cases. But after 40 seconds of execution the PDR increases proportionally to the increase in the number of OLSR-SOM represented by OLSR-SOM1, OLSR-SOM2, OLSR-SOM3, OLSR-SOM4 and OLSR-SOM5. This latter shows a significant increase over the standard OLSR hence the OLSR-SOM is better than the OLSR-standard regarding the PDR parameter.
The Figure 3 shows that that the average throughput in the MANET network at the beginning of the simulation has the same value for the different cases and that during the learning step of the SOM neural network. After around 40 seconds of execution that the average throughput becomes increasingly important by increasing the number of OLSR-SOM nodes. the figure also shows that the curves of the different cases of OLSR-SOM during the simulation time show higher values compared to the OLSR standard.
In the Figure 4, we note that the simulation from 0 second to about 50 second. The loss of packets has the same value for all the cases but from 60 to 135 second the loss of packets for standard OLSR begins to increase and exceeds the curves of OlSR-SOM at beyond 135 seconds. The decrease in packet loss is relevant when comparing the new approach in particular "OLSR-SOM5" against the standard OLSR.

Performance evaluation of OLSR-SOM against the DDOS attack
In order to evaluate the performance of the new OLSR-SOM MPRs selection approach against the DDoS attack, we consider the following scenario in which we applied OLSR-SOM on all nodes in the network with the presence of a DDOS attack launched by a given node. The simulation parameters are given in Table 2   In this second scenario, we applied OLSR-SOM on all nodes in the network with the presence of a DDOS attack launched by a given node. The obtained results are presented in Figures 6-9. On the one hand, the evolution of the PDR parameter for the OLSR standard compared to OLSR-DDOS illustrated in Figure  6(a), shows that the injection of DDOS attack into the network significantly reduced the PDR compared to standard OLSR. On the other hand, the evolution of PDR for OLSR-SOM compared to OLSR-DDOS illustrated in Figure 6(b) shows that OLSR-SOM increased the PDR compared to OLSR-DDOS by discarding nodes overloaded by DDOS and using the less overloaded MPR nodes to transmit messages in the network. So, the new approach also reacts against attacks to select the not overloaded MPR.
The evolution of the throughput parameter for the OLSR standard compared to OLSR-DDOS illustrated in Figure 7(a), shows that the DDOS attack overloading the network and blocking the functioning of the standard OLSR hence the low value of throughput parameter. However, the evolution of the throughput parameter for the OLSR-SOM compared to OLSR-DDOS illustrated in Figure 7(b) shows that the OLSR-SOM gives better throughput than OLSR-DDOS. The evolution of packet loss rate parameter for the OLSR standard compared to OLSR-DDOS illustrated in Figure 8(a), shows that the flooding created by DDOS attack at the network level has blocked the transmission of packets and therefore a very high packet loss rate. However, we notice that the evolution of packet loss rate parameter for the OLSR-SOM compared to OLSR-DDOS illustrated in Figure 8(b) shows that the OLSR-SOM has reduced this rate by avoiding the selection of weak MPRs. Concerning the evolution of the energy consumption parameter, the Figure 9 shows a similarity between the values given by the curves of the different protocols studied, namely OLSR, OLSR-DDOS and OLSR-SOM. The difference between them is that the OLSR-SOM compared to OLSR standard and OLSR-DDOS protocol has presented an importants results for the other parameters (PDR, throughputs and packet loss). in contrary, the OLSR-DDOS is an attack that causing the bad values for PDR, throughputs and packet loss parameters hence the important role of our new approach.

CONCLUSION
In this article, we have tried to improve both the quality and the security of the OLSR protocol and precisely the method of selecting MPRs nodes. To achieve this goal, we have integrated several parameters in the calculation of the MPRs noeds using the artificial neural network approach called SOM which makes it possible to classify the collected data into a defined number of classes, the application of a filter on these classes will make it possible to determine the strong and weak MPRs nodes and therefore the use of the strong MPRs in a final list in routing process. The obtained results from series of simulations using NS3 platform has proved the high performance of OLSR-SOM approach compared to standard OLSR in terms of both quality and security, especially with the observed increase in the throughput and the PDR, and with the decrease in the packet loss, whereas the energy consumption has kept the same level.