Mixture Localization-Based Outliers Models for securing Data Migration in Cloud Centers

Osama Al-Kadi, Nour Moustafa, Benjamin P. Turnbull, Kim-Kwang Raymond ChooIEEE Access 2019 · DOI · OpenReview

Introduction

Cloud computing environments rely on the dynamic movement of Virtual Machines (VMs) to optimize resource allocation, perform maintenance, and ensure high availability. This process, known as live migration, involves transferring the active state and memory of a VM from one physical host to another. While essential for operational efficiency, live migration introduces significant security risks. Data in transit can be targeted by both external attackers and malicious insiders, making the migration process a critical bottleneck for cloud security.

The Security Bottleneck in VM Migration

During migration, the memory contents of a VM are exposed as they traverse the network. Attackers can exploit this window of vulnerability to perform data interception, malicious code injection, or "VM escape" maneuvers. Existing security solutions, such as traditional Network Intrusion Detection Systems (NIDS), often fail to provide the granular visibility required to distinguish between legitimate high-volume traffic and sophisticated, low-volume attacks during these transitions.

Comparison of Pre-copy and Post-copy VM Migration Stages

As illustrated in the migration stage comparison, both pre-copy and post-copy methods involve critical "downtime periods" and "stop and copy" stages where the VM state is particularly vulnerable to interference. The MLO system is designed to monitor these specific transitions for anomalies.

The Mixture Localization-Based Outliers (MLO) System

To address these challenges, this research proposes the Mixture Localization-based Outliers (MLO) system. MLO is a collaborative anomaly detection framework that integrates Host-based IDS (HIDS), Network-based IDS (NIDS), and Virtual Trusted Platform Modules (vTPM) to ensure cryptographic integrity.

The MLO detection pipeline consists of three core phases:

  1. Pre-processing and Dimensionality Reduction: Raw network traffic is cleaned, and Principal Component Analysis (PCA) is applied to extract the most relevant features. This ensures the system can operate in real-time without excessive computational overhead.
  2. Gaussian Mixture Modeling (GMM): Unlike traditional methods that assume a single "normal" distribution, MLO uses GMM to create a multi-modal profile of network behavior. This flexibility allows the system to accurately model complex, high-variance traffic patterns common in data centers.
  3. Local Outlier Factor (LOF) Analysis: The system applies an LOF function to the probability densities generated by the GMM. By measuring the local density deviation of data points relative to their neighbors, MLO can identify subtle anomalies that might bypass global threshold-based systems.
Interaction Model between User, Service, and Cloud Environments

The collaborative nature of this system ensures that interactions between users, services, and the cloud infrastructure are monitored holistically, as illustrated in the interaction model above. This multi-layered approach is essential for detecting both external breaches and insider threats.

Experimental Results and Performance

The MLO system was rigorously tested against two industry-standard datasets: UNSW-NB15 and BoT-IoT. These datasets provide a realistic mix of normal traffic and modern attack vectors, including DDoS, worms, and reconnaissance activities.

The experimental results demonstrate superior performance compared to existing techniques like Triangle Area Nearest Neighbors (TANN) and Multivariate Correlation Analysis (MCA):

  • Detection Rate (DR): Reached 98.81% on the UNSW-NB15 dataset and 97.98% on the BoT-IoT dataset.
  • False Positive Rate (FPR): Maintained a low FPR, dropping to 1.26% as the model was refined with larger training samples.
  • Specific Attack Performance: The system was exceptionally effective against Denial of Service (DoS) and Distributed Denial of Service (DDoS) attacks, achieving detection rates above 98%.

Conclusion and Future Directions

The MLO framework provides a robust, adaptive shield for cloud data migration. By focusing on statistical flow features rather than deep packet inspection, it remains effective even against encrypted traffic while minimizing overhead. Future research will focus on evaluating the system's impact on Service Level Agreements (SLAs), specifically regarding migration latency and downtime in heterogeneous, large-scale production cloud deployments.

Summary updated

Abstract

The cloud computing paradigm is changing how businesses operate, providing greater efficiency, tolerance, elasticity and flexibility in computing workloads. Underpinning these changes are multiple data centers, operated by different entities and distributed globally. Despite these benefits, cloud computing presents new classes of cyber-attack, opportunities to attack and processes to subvert. One of the primary strategies to defend against cyber-attacks is the migration process. A secure Virtual Machine (VM) migration is essential to safeguard cloud data centers against insider and outsider attacks. In this paper, we propose a collaborative anomaly detection system for discovering insider and outsider attacks from cloud systems and their migration process. The proposed system is named Mixture Localization-based Outliers (MLO) and utilizes Gaussian-mixture models for fitting network data and a local outlier factor function for discovering abnormal patterns in network traffic data. In order to validate the effectiveness of the models, the datasets of UNSW-NB15 and BoT-IoT are employed. The experimental results have revealed the high performance of the proposed system compared with several peer anomaly detection techniques.
BibTeX
@article{alkadi2019mixture,
  author = {Osama Al-Kadi and Nour Moustafa and Benjamin P. Turnbull and Kim-Kwang Raymond Choo},
  title = {{Mixture Localization-Based Outliers Models for securing Data Migration in Cloud Centers}},
  journal = {IEEE Access},
  year = {2019},
  doi = {10.1109/access.2019.2935142},
}
Figure