Abstract
With the development of large-scale data, users increasingly need to store data in distributed storage systems because single computers cannot hold massive datasets. However, users often lack control over data access rules, making transparent security management in distributed networks a significant challenge. This paper proposes a distributed security storage model designed to handle high concurrency and complex data management. The Transparent Secure Cloud Storage System (TSCSS) provides confidentiality, integrity protection, and access control while maintaining I/O performance loss at less than 5%.
Introduction
Cloud storage offers scalability but often requires users to trust third-party providers. Internal threats and privilege abuse account for a significant portion of data breaches. Existing encrypted file systems frequently struggle with key management complexity, integrity verification, and efficient access revocation. TSCSS addresses these issues by implementing a "stacked" encryption file system that maintains end-to-end security on untrusted platforms without altering the underlying storage architecture.
System Architecture
The TSCSS architecture is divided into three main components: the Control Server (CS), the TSCSS Client, and the Storage Server (SS).
The system is built as a user-mode file system using FUSE (Filesystem in Userspace). This allows it to sit transparently atop existing systems like NFS or Ext3. The Control Server manages master keys and identity authentication via X.509 certificates, while the client handles data encryption, decryption, and integrity protection.
Security Mechanisms
Three-Layer Key Hierarchy
To manage keys efficiently at scale, TSCSS employs a hierarchical approach:
- File Block Key: Each 64KB data block has a unique key derived from its hash and offset.
- Metadata Lock: These keys are stored in a metadata file, protected by Key Locks (KL) and Signature Locks (SL).
- Master Keys: The Control Server stores only two master symmetric keys used to wrap/unwrap the metadata locks.
Integrity and Revocation
Integrity is maintained using Size Balanced Trees (SBT) to store block hashes. SBTs are preferred over Merkle trees because they remain more balanced during frequent updates, optimizing performance for dynamic data. Additionally, the system uses "lazy revocation"—when a user's access is removed, data is only re-encrypted during the next modification, preventing system paralysis.
Performance and Results
Evaluations using IOzone and Bonnie++ demonstrate the system's efficiency:
- Large-Scale Performance: In cluster environments, TSCSS achieved 95% of native NFSv4 performance.
- Low Latency: Permission updates (sharing or upgrading privileges) take approximately 2.5ms.
- Small File Penalty: For files around 512KB, performance drops by 43-50% due to the high metadata-to-data processing ratio.
- I/O Overhead: Sequential writes experienced a 22.2% increase in I/O operations due to simultaneous data and SBT metadata updates.
Conclusion
TSCSS provides a robust model for securing large-scale distributed data. While it excels in high-concurrency environments with large files, future work is needed to optimize the metadata overhead for small file operations and further distribute the Control Server for enhanced scalability.
Summary updated