Critical Workload Deployment in Public Clouds with Guaranteed Security Levels and Optimized Resource Usage and Energy Cost

Abstract

It is a common practice that public clouds adopt Virtual Machine (VM) multiplexing to improve resource usage and energy consumption. However, packing multiple VMs of different security requirements into a single hypervisor gives rise to major cybersecurity issues, such as VM to VM Interdependency-based cybersecurity (ICS) risks. For example, the chances of successfully compromising a secure Critical VM (CVM) are very high when an attacker compromises the hosting hypervisor after a successful attack on one of its less secure, non-critical VMs (NVMs). In this paper, we study how to securely and efficiently collocate CVMs with NVMs in public cloud clusters. Specifically, we model and analyze the ICS risks imposed on CVMs by NVMs using noncooperative game models involving two players, i.e., an attacker and a cloud provider. We then introduce a novel approach that can judiciously determine the allocation of VMs so that the ICS risks imposed on critical VMs are guaranteed to be minimized. Our experimental results show that our proposed algorithm can judiciously optimize the provider’s overall resource usage, energy consumption, and operational expense while minimizing the potential security loss given a successful attack on any VM.