DECENT: A Decentralized Architecture for Enforcing Privacy in Online Social Networks

A multitude of privacy breaches, both accidental and malicious, have prompted users to distrust centralized providers of online social networks (OSNs) and investigate decentralized solutions. We examine the design of a fully decentralized (peer-to-peer) OSN, with a special focus on privacy and security. In particular, we wish to protect the confidentiality, integrity, and availability of user content and the privacy of user relationships. We propose DECENT, an architecture for OSNs that uses a distributed hash table to store user data, and features cryptographic protections for confidentiality and integrity, as well as support for flexible attribute policies and fast revocation. DECENT ensures that neither data nor social relationships are visible to unauthorized users and provides availability through replication and authentication of updates. We evaluate DECENT through simulation and experiments on the PlanetLab network and show that DECENT is able to replicate the main functionality of current centralized OSNs with manageable overhead.

💡 Research Summary

The paper presents DECENT, a fully decentralized peer‑to‑peer architecture for online social networks that aims to protect user privacy while preserving the core functionality of modern OSNs. The authors begin by outlining the privacy shortcomings of centralized platforms, where the service provider has unrestricted access to user data and can change privacy policies arbitrarily. To address these issues, DECENT introduces a hierarchical object‑oriented data model in which every piece of content—profiles, posts, comments, albums, etc.—is represented as a container object that may reference other objects. Each object carries three independent access policies: a read policy (attribute‑based), a write policy (identity‑based, usually the owner), and an append policy (attribute‑based for comments).

Confidentiality is enforced using Attribute‑Based Encryption (ABE). Users act as key authorities, issuing decryption keys tied to attribute sets such as “friend”, “family”, or “coworker”. The actual content is encrypted with a random symmetric key; this key is then encrypted with ABE according to the object’s read policy and stored in the object’s reference rather than the object itself, preventing storage nodes from learning the policy. To support immediate revocation, the authors adopt the EASiER scheme, extending it with threshold secret sharing to distribute the proxy function across multiple random nodes and with attribute delegation to enable “friend‑of‑a‑friend” style policies.

Objects are stored in a Distributed Hash Table (DHT) such as Pastry or Kademlia. Each object is identified by a random objID and replicated across several neighboring nodes to ensure availability despite churn or malicious behavior. Metadata includes a version number and a write‑authentication public key (WAPK); any write or delete request must be signed with the corresponding secret key, allowing storage nodes to reject unauthorized modifications. The write policy’s public key is also part of the reference, enabling readers to verify that updates come from the legitimate owner.

The prototype implements core wall and news‑feed functionalities and is evaluated using the FreePastry simulator and a PlanetLab deployment of Kademlia. Results show that lookup latency is roughly 1.5–2× that of a centralized service, which the authors deem acceptable given the added security guarantees. Replication and versioning successfully preserve data integrity and freshness even when up to 20 % of nodes behave maliciously. The main performance overhead stems from ABE operations, which take a few hundred milliseconds on contemporary hardware.

In the discussion, the authors acknowledge limitations such as the computational cost of ABE, potential bottlenecks at proxy nodes during revocation, and the lack of DRM‑style controls over re‑sharing of object references. Future work includes exploring more efficient ABE constructions, load‑balancing strategies for proxies, and mechanisms to deter unauthorized redistribution of references.

Overall, DECENT demonstrates that by combining an object‑oriented data model, attribute‑based cryptography with fast revocation, and robust DHT replication, it is possible to build a decentralized social network that simultaneously offers confidentiality, integrity, availability, fine‑grained access control, and relationship privacy—features that existing decentralized designs typically address only partially.