Detecting Colluding Attackers in Distributed Grid Systems

Abstract

Distributed grid systems offer possible benefits in terms of fast computation of tasks. This is accompanied by potential drawbacks due to their openness, the heterogeneity of participants, and the unpredictability of agent behaviour, since agents have to be considered as black-boxes. The utilisation of technical trust within adaptive collaboration strategies has been shown to counter negative effects caused by these characteristics. A major challenge in this context is the presence of colluding attackers that try to exploit or damage the system in a coordinated fashion. Therefore, this paper presents a novel approach to detect and isolate such colluding attackers. The concept is based on observations of interaction patterns and derives a classification of agent communities. Within the evaluation, we demonstrate the benefit of the approach and highlight the highly reliable classification.