Modelisation des attaquants et de la reputation dans les reseaux ad hoc mobiles.

摘要

Mobile ad-hoc networks (MANET) are networks where nodes are users, servers and routers. This structure causes several security problems such as those caused by node greeniness. Nodes can drop packets which they should normally forward. Node reputation is used to identify and eliminate malicious nodes and prevent them from deteriorating network Quality of Service (QoS). However, performance analysis of reputation-based counter-measures (RBCM) is often flawed. The attackers, the environment and the performance parameters are not always well defined and there is no analytical model that can be used to compute the efficiency of RBCM.; The main objective of this Master's thesis is to define an analytical model that describes the interaction between MANET environment, attackers and performance criterias such as false positive rate, false negative rate and packet drop rate. With this model, it would be possible to know exactly the precision of a RBCM.; Stochastic processes are used to model node connectivity, traffic generation, transmissions between nodes, packets losses in order to compute quantitative metrics. Two main hypotheses are used. First, traffic changes very slowly. Second, the packet inter-arrival time distribution is exponential. Some of the stochastic processes are composed of nonlinear equations, which means that stochastic simulation methods must be used to resolve the model.; This model has been partly implemented to simulate a MANET using the DSR routing protocol without any RBCM. Performance analysis has been done by comparing the model results and those obtained with Qualnet, a commercial network simulator. The Kolmogorov-Smirnov statistical test demonstrates that the model cannot reproduce Qualnet results. There are big differences between the values for the probability to drop packets calculated with Qualnet and those calculated with the model. Special cases tested with Qualnet show that the inter-arrival time between frames that nodes must transmit is not exponentially distributed. This could explain the differences between the model results and those from Qualnet.; This is the first model that describes interactions between attackers and the environment. The other models only describe the environment or the attackers. However, it cannot be used to validate RBCM performance. In future work, it would be important to improve collision modeling in order to increase model precision. Only some model aspects were validated in this Master's thesis. It will thus be important to validate all aspects in future work. We give several lessons learned to help other researchers validate RBCM.