Knocking on IP: Unveiling Websites through Cache-Aware Fingerprinting

As user privacy becomes increasingly critical in the digital landscape, traditional methods of website fingerprinting (WF) face significant challenges, particularly in caching scenarios. Existing WF studies are limited by the assumption of disabled caching. Recently, only a few have explored how to address the potential problem of model performance degradation in caching scenarios, but what they propose have limitations in terms of transferability and practicality in existing conditions. However, we propose a deep learning method based solely on IP fingerprints, which is more resilient to network fluctuations than traditional features and is able to extract effective local features from the fluctuating sequences. Furthermore, when tested on simulated cached datasets, our method achieves a remarkable stability with an accuracy of 96.10%. These findings underscore our model’s suitability for identifying websites amidst mixed-length sequences in caching contexts. The solution we provide not only improves the accuracy of identification, but also reveals the impact of caching mechanisms on privacy. Meanwhile, our comprehensive dataset and innovative methods pave the way for future research in user privacy and WF.