Boosting Attack Detection Capabilities in Multi-Tenant Distributed Systems via Meta-Ensemble Classifiers and Weighted Averaging

Abstract

Multi-tenant distributed systems are a great way to share resources and scale performance, but they come with their share of security issues due to the introduction of more than one client on the cloud. Sharing of hardware and software resources among tenants gives rise to vulnerabilities and malicious tenants can misuse this shared data to launch attacks on other tenants. While efforts have traditionally focused on building secure network architectures, it is impossible to create a completely secure system due to its open-ended nature. This paper explores ways to detect malicious tenants on the cloud using machine learning algorithms. This paper proposes an ensemble-based meta-classifier to predict the probability of attack instantiation based on certain system parameter values. Additionally, this paper creates a dataset for analysis purposes and address the class imbalance problem often found in this domain where attack instances are rare. Satisfactory results were produced to distinguish between non attach and attack instances.