NET-SA: An Efficient Secure Aggregation Architecture Based on In-Network Computing

📅 2025-01-02
📈 Citations: 0
Influential: 0
📄 PDF

career value

220K/year
🤖 AI Summary
To address gradient leakage, high communication overhead, and inefficient secure aggregation caused by dynamic client dropout in privacy-preserving federated learning, this paper proposes an in-network computing–based secure aggregation architecture. Our method jointly designs homomorphic pseudorandom masking—using client-local seeds—with P4-programmable switches (Intel Tofino), enabling seed aggregation directly in the network fabric. This eliminates the need for key agreement and threshold secret sharing, thereby overcoming traditional threshold-based fault-tolerance constraints. The approach supports graceful handling of client dropouts without requiring a trusted third party. Experiments demonstrate that, compared to state-of-the-art schemes, our solution achieves up to 77× reduction in end-to-end runtime and cuts total client communication cost by 50%. We validate efficiency and practicality on both real server clusters and the Tofino hardware platform.

Technology Category

Application Category

📝 Abstract
Privacy-preserving machine learning (PPML) enables clients to collaboratively train deep learning models without sharing private datasets, but faces privacy leakage risks due to gradient leakage attacks. Prevailing methods leverage secure aggregation strategies to enhance PPML, where clients leverage masks and secret sharing to further protect gradient data while tolerating participant dropouts. These methods, however, require frequent inter-client communication to negotiate keys and perform secret sharing, leading to substantial communication overhead. To tackle this issue, we propose NET-SA, an efficient secure aggregation architecture for PPML based on in-network computing. NET-SA employs seed homomorphic pseudorandom generators for local gradient masking and utilizes programmable switches for seed aggregation. Accurate and secure gradient aggregation is then performed on the central server based on masked gradients and aggregated seeds. This design effectively reduces communication overhead due to eliminating the communication-intensive phases of seed agreement and secret sharing, with enhanced dropout tolerance due to overcoming the threshold limit of secret sharing. Extensive experiments on server clusters and Intel Tofino programmable switch demonstrate that NET-SA achieves up to 77x and 12x enhancements in runtime and 2x decrease in total client communication cost compared with state-of-the-art methods.
Problem

Research questions and friction points this paper is trying to address.

Distributed Learning
Privacy Protection
Communication Efficiency
Innovation

Methods, ideas, or system contributions that make the work stand out.

NET-SA
in-network computing
privacy-preserving aggregation
🔎 Similar Papers
No similar papers found.
Q
Qingqing Ren
Institute of Computing Technology, Chinese Academy of Sciences; University of Chinese Academy of Sciences
W
Wen Wang
Institute of Computing Technology, Chinese Academy of Sciences; University of Chinese Academy of Sciences
S
Shuyong Zhu
Institute of Computing Technology, Chinese Academy of Sciences; University of Chinese Academy of Sciences
Z
Zhiyuan Wu
Institute of Computing Technology, Chinese Academy of Sciences; University of Chinese Academy of Sciences
Y
Yujun Zhang
Institute of Computing Technology, Chinese Academy of Sciences; University of Chinese Academy of Sciences