Secure Collaborative Decision-Making Among Distributed Cloud Brokers Using Privacy-Preserving Federated Learning
TL;DR
Federated learning-based SecureBroker-FL is proposed, introducing 3 key aspects: A Differential Privacy enhanced Gradient Encryption (DPGE) protocol, an Adaptive Trust scoring (ATS) mechanism, and Hierarchical Secure Aggregation (HSA).
Abstract
Cloud brokers play a vital role in coordinating resource-allocation decisions across a heterogeneous cloud environment. Centralised brokerage approaches are prone to single points of failure and data exposure, expose sensitive data, and have limited interoperability. There is a need for distributed cloud brokerage that enhances collaboration by enabling decision-making without exposing raw workload specifications. A shared knowledge repository in collaborative learning introduces data sovereignty and adversarial risks. This paper proposes Federated learning-based SecureBroker-FL, introducing 3 key aspects: (i) A Differential Privacy enhanced Gradient Encryption (DPGE) protocol. (ii) An Adaptive Trust scoring (ATS) mechanism, (iii) Hierarchical Secure Aggregation (HSA). The experiments are simulated and evaluated using the Google Cloud Trace 2019 and Alibaba Clustered Trace 2022 datasets, with five geographically distributed cloud brokers. It demonstrates that SecureBroker-FL achieves 94.7% decision accuracy. The framework withstands up to 40% malicious broker participation handling graceful degradation without catastrophic collapse and outperforms baseline approaches, including FedAvg, Krum, and FLTrust.