Defense Against Attacks in Self-service Cloud Using Reinforcement Learning

Abstract

Cloud computing offers various services which are analogous to traditional data centers. The on demand supply of resources make this model of utility computing as the platform for many web based services. However, security is always a major concern. This thesis proposes a new architecture called Self-service cloud computing with virtual shield (VS) to secure the entire cloud environment. When a malicious attack is predicated, the Virtual shield (VS) dynamically changes the configurations of the client virtual machines (VM) using a reinforcement learning mechanism to achieve the required security. The system may be dynamically modified in response to changes in system configuration, state, and/or workload. The reward values generated during the learning process determines the reconfiguration of the client. Simulation results show that the dynamic reconfiguration of virtual machines when anticipated to confront an attack, diminishes the likelihood of an attack and secures the cloud virtual machines.