Content Distribution Networks consist of edge data centers that cache content in different geographical locations in order to provide redundancy and better content distribution performance. With the increase of real time data such as video streaming and the increasing of users requesting content, CDNs which are relying on dedicated servers do not scale well. Peer-assisted CDNs (pCDN) solve this problem by leveraging peers that received content to behave as relay of content to peers close geographically. pCDNs turn peers into caching nodes and make it cheaper for CDNs to scale and achieve better performances. If the pCDN protocol is developed naively and given the P2P nature of pCDNs, peers delivering and requesting data in the pCDN P2P network may be able to gather information that correlates IP addresses with online accessed content. This information can pose a critical threat to online privacy. The goal of this work is to perform inference attacks on deployed peer-assisted CDNs.
It is relatively easy and cheap to gather critical and private data of users using peer-assisted CDNs, which can be used to infer user behavior.
In order to test the hypothesis, we need to know which services are using a peer-assisted CDN network in order to perform the attack.
- database dump which correlate IPs with content and timestamps in different pCDN.
- scripts which automate gathering of metadata while serving/requesting content
- reach out: blog post, paper, talk
- Anonymity in Peer-assisted CDNs: Inference Attacks and Mitigation