Data protection: Combining fragmentation, encryption, and dispersion

Gérard Memmi, Katarzyna Kapusta, Han Qiu

TL;DR

This paper proposes a comprehensive data protection system (FEDS) that combines fragmentation, encryption, and dispersion to secure data throughout its lifecycle. It introduces novel selective encryption methods for images and explores fragmentation techniques for structured data, aiming for a balance of security, performance, and cost-effectiveness, particularly leveraging GPU acceleration.

What Problem Does This Paper Address?

In today's landscape of increasing data volumes and sophisticated cyber threats, relying solely on encryption for data protection is insufficient. Attacks can steal, alter, or destroy private information, and encryption itself can be computationally expensive and complex to manage, especially key management. The paper addresses the need for a more robust, cost-effective, and scalable approach to data protection that extends beyond simple encryption. It tackles the challenge of securing data across its entire lifecycle – from creation to storage and transmission – in diverse environments, including cloud platforms and resource-constrained devices.

Methods

The paper presents a multi-faceted approach:

  • Fragmentation: Data is broken down into smaller pieces (fragments) to reduce the attack surface and allow for differential protection. This includes:
  • Selective Fragmentation: Dividing data into a confidential (private) and a non-confidential (public) part. For images, this involves transforming the image into the frequency domain (using DCT or DWT) and encrypting only the most significant coefficients.
  • Non-Selective Fragmentation: Dividing data into fragments without regard for its structure or semantic meaning, often for resilience and dispersion.
  • Encryption: The confidential fragments are encrypted using robust algorithms like AES. The paper explores selective encryption for images, where only a portion of the data is encrypted, aiming for performance gains.
  • Dispersion: Fragments are distributed across multiple physical machines or storage locations to prevent an attacker from obtaining a complete picture of the data, even if they compromise a single machine.
  • GPU Acceleration: The paper heavily leverages Graphics Processing Units (GPUs) for computationally intensive tasks like DCT and DWT, significantly speeding up processing compared to traditional CPUs.
  • FEDS (Fragmentation, Encryption, Dispersion System): A conceptual framework that integrates these three components to provide end-to-end data protection.

Datasets

  • Bitmap Images: The paper uses standard bitmap images, including "Lenna," "Barbara," and "baboon," for evaluating image encryption techniques.
  • ITEA2 CAP Project Data: Data from La Poste (French Post Office) was used for scalability experiments.
  • Text Data: Used for evaluating fragmentation and encryption methods on non-image data.

No new benchmarks were introduced; instead, the paper compares its methods against existing standards like AES.

Key Insights and Ideas

  • Beyond Encryption: The paper strongly advocates for a layered security approach that combines fragmentation and dispersion with encryption, arguing that this is more effective than relying on encryption alone.
  • Selective Encryption for Performance: Encrypting only the most critical parts of data (e.g., low-frequency coefficients in images) can drastically improve performance without significantly compromising security, especially on resource-constrained devices.
  • GPU as a Game-Changer: The significant speedups achieved by using GPUs for image processing tasks like DCT and DWT demonstrate their crucial role in making advanced data protection techniques practical.
  • Agnostic Approach: The developed methods are designed to be agnostic to data type and structure, making them broadly applicable across various systems and data formats.
  • Multilevel Security: The concept of adapting storage architecture and protection levels to the confidentiality of data is crucial for cost-effectiveness and efficient resource utilization.

Key Results

  • Selective Encryption Speedup: The proposed selective encryption methods using DCT and DWT with GPU acceleration were shown to be significantly faster than full image encryption with AES, achieving speedups of 2-8 times on laptops and comparable performance to AES on powerful desktops.
  • Robustness: Statistical analysis (histogram uniformity, correlation coefficients, entropy) demonstrated that the encrypted data exhibits high randomness and independence from the original data, indicating resistance to statistical attacks.
  • Data Protection: The strong level of protection for images effectively blurred visual characters, making it difficult to recover meaningful information from the public fragment.
  • Fragmentation Efficiency: The fragmentation techniques, especially when combined with GPU acceleration, offer a promising way to manage large datasets while maintaining security and resilience.

Limitations and Future Work

  • Key Management for Long-Term Storage: The paper acknowledges that for long-term storage, key management remains a challenge, as keys can be lost or become vulnerable to advances in cryptanalysis.
  • Formal Security Analysis: While statistical tests were performed, the authors suggest the need for more formal methods to compare different FEDS approaches and set optimal parameters.
  • Dispersion and Decoys: The effect of dispersion strategies and the use of decoys on security needs further investigation.
  • Agnosticism Testing: More experiments are needed to test the methods with diverse data formats (HD images, audio, video) and use cases.
  • General Architecture: The ultimate goal is to design a general architecture that is cost-effective, performant, and resilient, potentially by further optimizing parallel processing and considering data separation based on confidentiality levels.

Summary updated

Abstract

In this paper, we first present a general description of what is a fragmentation system including a number of high level requirements. Then, we focus on fragmentation of two kinds of data. First, a bitmap image is split in two parts a public one and a private one. We describe two processes and address the question of performance. Then, we survey works on fragmentation of text: in a brute force manner then considering data stored in a structured database.
BibTeX
@inproceedings{memmi2015data,
  author = {Gérard Memmi and Katarzyna Kapusta and Han Qiu},
  title = {{Data protection: Combining fragmentation, encryption, and dispersion}},
  booktitle = {SSIC},
  year = {2015},
  doi = {10.1109/ssic.2015.7245680},
}

Loading related content...

Figure