Article Preview
Top1. Introduction
A critical MANET is a MANET that contains one or more critical node. By definition a critical node is a node whose malicious behaviour interrupts the underlying MANET (A.Karygiannis et al, 2006) (Shivashankar et al, 2012). Consequently such network will fail to achieve its mission which may be also critical in military or disaster recovery applications. Typically, they consist of a set of servers where one of them is elected as a leader and a set of core-based clients where that core is materialized by a gateway. The entire security of such MANETs is complex, labor and multi-disciplinary. In fact, there are some papers that study the consensus of the network servers only (M.Henrique et al, 2012) (M.Toulouse et al, 2016) (A.Geetha et al, 2016), however, that approach does not take into consideration the failure of some clients, consequently results in higher false positive values than the ground truth. On the other hand, some other researchers concentrate only (A.Mitrokotsa et al, 2013) (A.Khannous et al, 2014) (M.Toulouse et al, 2015) on the MANET clients, however, that approach does not take into account the Byzantine servers, and it results in higher false positive values than the reality. Unfortunately, there is no a single model that covers simultaneously the two sides (servers and clients) of a critical MANET.
This paper presents CSMCSM or shortly (CSM)2 as a formal model that studies for the first time the security of both sides of critical MANETs. For such model the following aspects are pointed out.
- 1.
iThe Client Server Model (CSM) is a distributed system model that works by describing how servers provide services, management and control to service requesters that are represented as clients. In MANETs secured communication is most important, because of their wireless medium, resource limitation and dynamic topology. They are more prone to dangerous security attacks that come from either insiders (Byzantine) or outsiders (traditional).
- 2.
(CSM)2 is Raft-based that employs the Raft (D.Ongaro et al, 2013) as a secure consensus algorithm.
- 3.
In designing Raft, researchers Ongaro and Ousterhout applied specific techniques to improve understandability of the formal consensus concept. From the names of the two researchers the algorithm name (Tangaroa) is formed (C.Copeland et al, 2014) has presented Tangaroa as an extension of Raft. The Byzantine fault tolerant, BFT, approach given by Tangaroa is used to mitigate higher false positive values.
- 4.
The decision trees generated by C4.5 can be used for classification. C4.5 which is often referred to as a statistical classifier. In 2011, authors of the Weka(R.Bouckaert et al, 2013) machine learning software described the C4.5 algorithm as “a landmark decision tree program”. A C4.5 tree classifier (H.Chauhan et al,2013) is integrated with Tangaroa (that works as a BFT system) to solve the problem of higher false positive values. Here the failed consensus cases, only, are considered suspected and are passed to the decision tree. Thus the C4.5 tree classifier extends Tangaroa to allow more accurate results.
- 5.
As Application Needs MANET Simulation “ANMS” software is the heart of (CSM)2. It simulates the typical attacks integrates with Tangaroa to enable it to avoid high false positives and it passes the attack features to the decision tree classifier. In other words ANMS acts as an umbrella for both server and client sides, consequently integrates smoothly BFT Tangaroa and C4.5 (J48 tree classifier).
- 6.
ANMS has the advantage of MANET energy saving since it does not pass all the operational cases to the classifier to test them, only the cases that failed in reaching consensus are passed to be categorized by the classifier.
(CSM)2 is prototyped and its performance is evaluated. The prototype has the following contributions:
- 1.
Handling comprehensive security of both servers and clients i.e. of known and unknown MANET participants.
- 2.
Capability of energy saving to elongate the life time of the underlying MANET.
- 3.
High accuracy by minimizing both false positives and false negatives.