188宝金博页面版

Evolving Attackers against Wireless Sensor NetworksKinga MrugalaDepartment of ComputerScienceUCL, Londonkinga.mrugala.11@ucl.ac.ukNilufer TuptukDepartment of ComputerScienceUCL, Londonn.tuptuk@cs.ucl.ac.ukStephen HailesDepartment of ComputerScienceUCL, Londons.hailes@cs.ucl.ac.ukABSTRACTRecent technological improvements in wireless communica-tion and electronics have enabled the development of small,low-cost wireless sensor nodes, capable of monitoring every-thing from human health to the performance of the electric-ity grid. A natural consequence is a desire to secure systemscontaining these nodes. Unfortunately, proving that systemsare secure is beyond the current state of the art, and test-ing for security is problematic: test cases often miss attacksthat have never previously been seen. In this paper, we useGenetic Programming (GP) to create attacks against Inter-net of Things devices, to help identify vulnerabilities beforesystems are attacked for real. To assess the ef f ectivenessof each attacker, we used it against a wireless sensor net-work (WSN) with publish-subscribe communications, pro-tected by a literature artif i cial immune intrusion detectionsystem (IDS). The GP attackers succeeded in suppressingsignif i cantly more legitimate messages than a hand-codedattacker, whilst decreasing the likelihood of detection. Asa consequence, it was possible to tune the IDS, improvingits performance. Whilst these results are preliminary, theydemonstrate GP holds signif i cant potential for improvingthe protection of systems with large attack spaces.CCS Concepts•Security and privacy → Artif i cial immune systems;Mobile and wireless security; •Computing method-ologies → Genetic programming; •Networks → Sensornetworks;KeywordsEvolutionary Computation, Wireless Network Attacks, Rout-ing Protocols1. INTRODUCTIONThe idea of an Internet of Things (IoT) in which objectsof all kinds are networked together, has become so pervasivePermission to make digital or hard copies of part or all of this work for personal orclassroom use is granted without fee provided that copies are not made or distributedfor prof i t or commercial advantage and that copies bear this notice and the full citationon the f i rst page. Copyrights for third-party components of this work must be honored.For all other uses, contact the owner/author(s).GECCO’16 Companion July 20-24, 2016, Denver, CO, USAc ? 2016 Copyright held by the owner/author(s).ACM ISBN 978-1-4503-4323-7/16/07.DOI: http://dx.doi.org/10.1145/2908961.2908974that it is at the heart of much current academic interest [6]as well as the commercial strategies of large companies [1].Naturally, this gives rise to security concerns, which becomeeven more important when one realises that IoT technolo-gies are regarded as an exciting near-term prospect for usein industrial control systems - so called Industry 4.0. Wepresent initial results that suggest that evolutionary com-puting techniques have the potential to revolutionise theway in which we engineer IoT systems by facilitating theautomated creation of attackers against which one can test(and so adapt) one’s defensive strategies.Previous studies [4] have used GP to evolve attacks againstwired networks, but as far as we know, there are no studiesthat have investigated using GP to evolve attacks againstwireless sensor networks. In this paper, we explore a limitedproblem to demonstrate the feasibility of evolving attackers.We study a WSN protected with an artif i cial immune system(AIS) devised by Wallenta et al. [5]. We evolve attackersagainst it, and demonstrate their ef f ectiveness in comparisonto the strategies that are used in evaluating the ef f ectivenessof the original AIS.2. PRELIMINARIESIn this paper, we use the Sensor Network Based Artif i cialImmune System (SNAIS) [5], an intrusion detection system,based on a Dendritic Cell (DC) Algorithm [2]. SNAIS is de-signed to operate in the context of a WSN that uses directeddif f usion [3]. Directed dif f usion is a routing algorithm usedto gather data from a large number of nodes and routing itto nodes that request the data. SNAIS operates in a simi-lar manner to the human immune system, where signals aregathered and then used to decide on future actions.For this study, we consider attacks based on the interestcache poisoning attack described in [5]. In an interest cachepoisoning attack, an attacker attempts to inject bogus in-terest entries into the interest cache of a node. Under thedirected dif f usion protocol, the node chooses to where tosend a data message using the interest cache entries. If le-gitimate entries can be forcefully eliminated from the cacheby an attacker, nodes will drop legitimate data packets.3. EXPERIMENTSGenetic programming was used as the evolutionary ap-proach for attack generation. To explore the problem, twof i tness functions were used: the f i rst focuses on suppressingthe number of packets delivered from source to sink; thesecond focuses on trying to lower the precision and recallof SNAIS. Moreover, the attackers were also evolved on two107