Defender Vs Attacker Security Game Model for an Optimal Solution to Co-resident DoS Attack in Cloud

Abstract

Virtual Machines (VM) are considered as the fundamental components to cloud computing systems. Though VMs provide efficient computing resources, they are also exposed to several security threats. While some threats are easy to block, some attacks such as co-resident attacks are much harder even to detect. This paper proposes Defender Vs Attacker Security Game Model otherwise called Two-Player security game approach based defense mechanism for minimizing the Co-resistance DOS attacks by making it hard for intruders to initiate attacks. The proposed defense mechanism first analyzes the attacker behavior difference between attacker and normal users under PSSF VM allocation policy. Then the clustering analysis is performed by EDBSCAN (Enhanced Density-based Spatial Clustering of Applications with Noise). The partial labeling is done depending on the clustering algorithm to partially distinguish the users as legal or malicious. Then the semi-supervised learning using Deterministic Annealing Semi-supervised SVM (DAS3VM) optimized by branch and bounds method is done to classify the nodes. Once the user accounts are classified, the two-player security game approach is utilized to increase the cost of launching new VMs thus minimizing the probability of initiating co-resident DOS attack.