.png){kind=icon}
Return
CompactDB: Compact and Secure Collaborative Relational Data Analytics
.png){kind=icon}
PRE
AI
DOI：10.1109/tdsc.2026.3723679.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Modern data analytics workflows increasingly rely on joint datasets and collaborative relational analytics to support diverse applications. Because client data often contains sensitive information, it must be protected from direct sharing. Integrating outsourced computation with Secure Multi-Party Computation (MPC) has emerged as a promising paradigm for secure collaborative relational analytics, but existing implementations face fundamental efficiency challenges due to the inherently communication-intensive nature of MPC. To address this limitation, we propose CompactDB, a compact and secure framework for collaborative relational data analytics. Our key observation is that existing solutions typically rely on secure predicate evaluation protocols that produce non-compact output tables, because revealing evaluation results directly may lead to privacy leakage. To overcome this, we design a communication-efficient compaction protocol that securely filters out non-matching records during predicate evaluation, yielding significantly more compact output tables. Building on this idea, we further develop a communication-efficient operator for multi-predicate evaluation on low-precision data, and introduce a new scheduling strategy that systematically exploits our compaction protocol within complex query plans. We conduct comprehensive experiments to evaluate CompactDB, demonstrating performance improvements of $6\times$ to $100\times$ over state-of-the-art solutions on TPC-H Q4 and Q13. In addition, we provide a theoretical analysis proving the security guarantees achieved by CompactDB.
Keywords:
Secret Shared Database
Secure Compaction
Function Secret Sharing
Privacy Enhancing Technology
Journal
IF:
7.5
Papers:
2.5K
Citations:
9.6K
Organization
Cited Papers
No cited papers available