Towards Social Profile Based Overlays

Towards Social Profile Based Overlays

David Isaac Wolinsky, Pierre St. Juste, P. Oscar Boykin, Renato Figueiredo
University of Florida

Towards Social Profile Based Overlays

David Isaac Wolinsky, Pierre St. Juste, P. Oscar Boykin, Renato Figueiredo

University of Florida


Online social networking has quickly become one of the most common activities of Internet users. As social networks evolve, they encourage users to share more information, requiring the users, in turn, to place more trust into social networks. Peer-to-peer (P2P) overlays provide an environment that can return ownership of information, trust, and control to the users, away from centralized third-party social networks.

In this paper, we present a novel concept, social profile overlays, which enable users to share their profile only with trusted peers in a scalable, reliable, and private manner. Each user’s profile consists of a unique private, secure overlay, where members of that overlay have a friendship with the overlay owner. Profile data is made available without regard to the online state of the profile owner through the use of the profile overlay’s distributed data store. Privacy and security are enforced through the use of a public key infrastructure (PKI), where the role of certificate authority (CA) is handled by the overlay owner and each member of the overlay has a CA-signed certificate. All members of the social network join a common public or directory overlay facilitating friend discovery and bootstrap connections into profile overlays. We define interfaces and present tools that can be used to implement this system, as well as explore some of the challenges related to it.

1 Introduction

Online social networking has become pervasive in daily life, though as social networks grow so does the wealth of personal information that they store. Once information has been released on a social network, known as a user’s profile, the data and the user are at the mercy of the terms dictated by the social network infrastructure, which today is typically third-party, centrally owned. If the social network engages in activities disagreeable to the user, due to change of terms or opt-out programs not well understood by users such as recent issues with Facebook’s Beacon program [14], the options presented to the user are limited: to leave the social network (surrendering their identity and features provided by the social network), to accept the disagreeable activities, or to petition and hope that the social network changes its behavior.

As the use of social networking expands to become the primary way in which users communicate and express their identity amongst their peers, the users become more dependent on the policies of social network infrastructure owners. Recent work [3] explores the coupling between social networks and P2P systems as a means to return ownership to the users, noting that a social network made up of social links is inherently a P2P system with the aside that they are currently developed on top of centralized systems. In this paper, we extend this idea with focus on the topic of topology; that is, how to self-organize social profiles that leverage the benefits offered by a structured P2P overlay abstraction.

Structured P2P overlays provide a scalable, resilient, and self-managing platform for distributed applications. Structured overlays enable users to easily create their own decentralized systems for the purpose of data sharing, interactive activities, and other networking-enabled activities. In recent work [22], we implemented mechanisms that allow users to create and manage their own private overlays using a common public overlay to assist in discovery and NAT traversal. This prior work focuses on generic structured P2P private overlays; in this paper, we expand upon this approach with in-depth discussion on how to apply this technique to enable social network overlay profiles.

Social networks consist of users, each has a profile, a set of friends, and private messaging. The profile contains user’s personal information, status updates, and public conversations, similar to a message board. Friends are individuals trusted sufficiently by a user to view the user’s profile. Private messaging enables sending messages discretely between users without leaking the message to other members. Using this model, we describe how a public overlay can be used as a directory for finding friends and joining existing profile overlays. Each user has their own profile overlay, secured via a public key infrastructure (PKI) to which they are the certificate authority (CA). The profile overlay stores profile data in its distributed data store, supporting profile access in scalable mechanisms regardless of the profile owner’s online status. In this paper, we present the architecture of these overlays, as presented in Figure 1, and how they are used to find and befriend peers, and describe approaches to handling profile data and establishing initial connections to profile overlays.

Figure 1: An example social overlay network. Alice has a friendship with Bob and Carol, hence both are members of her profile overlay. Bob has a friendship with Alice and Dave but not Carol; hence Alice and Dave are members of his profile overlay, while Carol is not. Each peer has many overlay memberships but a single root represented by dashed lines in various shades of gray. For clarity, overlay shortcut connections are not shown.

The rest of this paper is organized as follows. Section 2 provides background and related work. Section 3 describes our multi-overlay approach, explaining how to map social networks onto structured P2P overlays. In Section 4, we explore some of the remaining challenges introduced by our approach. We conclude the paper in Section 5.

2 Background

In this section, we review structured P2P overlays and distributed, decentralized online social network approaches.

2.1 Structured P2P Overlays

Structured P2P systems provide distributed look-up services with guaranteed search time with a lower bound of , in contrast to unstructured systems, which rely on global knowledge/broadcasts, or stochastic techniques such as random walks [6]. Some examples of structured systems can be found in [18, 20, 12, 13, 16, 10]. In general, structured systems are able to make these guarantees by self-organizing a structured topology, such as a 2D ring or a hypercube.

In the overlay, each node is given a unique node ID drawn from a large address space. Each node ID must be unique otherwise address collisions will occur, which can prevent nodes from participating in the overlay. Furthermore, having the node IDs well distributed assists in providing better scalability as many shortcut selection algorithms depend on having node IDs uniformly distributed across the entire address space. Two approaches to ensure this behavior are to have each node use a cryptographically strong random number generator to generate the node ID, or to use a trusted third party generate node IDs and cryptographically sign them [7].

Overlay shortcuts enable efficient routing in ring-structured P2P systems. Different shortcut selection methods include: maintaining large tables without using connections and only verifying usability when routing messages [18, 13], maintaining a connection with a peer every set distance in the P2P address space [20], or using locations drawn from a harmonic distribution in the node address space [12].

Most structured P2P overlays support decentralized storage/look-up of information by mapping keys to specific node IDs in an overlay. At a minimum, the data is stored at the node ID either smaller or larger to the data’s node ID and for fault tolerance the data can be stored at other nodes. This sort of mapping and data storage is called a distributed hash table (DHT). DHTs provide the building blocks to form more complex distributed data stores as presented in Past [17] and Kosha [5].

In [8, 15], the authors discuss the concept of a single overlay supporting services through the use of additional overlays, which use the underlying overlay to assist in discovery. In [22], we presented a reference implementation of a multiple-overlay system that supports the use of a public overlay’s DHT to store currently active peers in the private overlays. The system allows users to create their own private overlays without having to create their own bootstrap network. During evaluation, using both simulated and real systems, the time for a single peer to join first the public and thereafter private overlays was small and grew logarithmically with network sizes. The real system was tested using a public overlay of 600 nodes on PlanetLab and a random distribution of peers in the private overlay. With point-to-point security links enabled in the private overlay, the time to connect was less than 22 seconds for all cases. In simulations, overlays with as many as 100,000 peers were evaluated. For the 100,000-peer overlay, regardless of the private overlay size and with security enabled, peers were able to connect to their private overlays in less than 48 seconds. In relation to this paper, these results can be interpreted such that the latency required for a single peer to from being completely disconnected from the social network to being fully connected to the directory and all profile overlays.

In addition, our system provides both relay-based and hole-punching NAT traversal [21] techniques and supports point-to-point PKI based security [22], forming a basis for the approach presented in this paper.

2.2 Peer-to-Peer Social Networks

In [4], a DHT provides the look-up service for storing meta data pertaining to a peer’s profile. Peers query the DHT for updated content from their friends by hashing their unique identifiers (e.g. friends’ email addresses). The retrieved meta data contains information for obtaining the profile data such as IP address and file version. Their work relies on a PKI system that provides identification, encryption, and access control. In contrast, our approach provides each user their own private overlay secured by point-to-point encryption and authentication amongst all peers in the profile overlay. The profile overlay provides a clean abstraction of access control, whereby once admitted to a private overlay, users can access a distributed data store which holds the contents of the owners profile.

[19] takes a different approach by depending on virtual individual servers (VIS) hosted on a cloud infrastructure such as Amazon EC2. Friends contact each other’s VIS directly for updates. A DHT is used as a directory for groups and interest-based searches. Their approach assumes bidirectional end-to-end connectivity between each VIS, where a profile is only available during the up time of the VIS. Because of the demands on network connectivity and up time, the approach assumes a cloud-hosted VIS and has difficulty being used on user-owned resources. Our approach enables users to avoid the need for all-to-all connectivity and constant up time through the use of NAT traversal support and the ability to store the profile in the overlay’s distributed data store.

The approach presented in [9] also uses a DHT for looking up a peer’s circle of friends. Once a node in the peer’s outermost circle is found, that node is used to route profile requests to the innermost circle which contains replicas of a peer’s profile. Trust is enabled through the use of an identification service contacted through the DHT. The circle of friends concept lacks the simplicity of the abstraction made in our approach, which can easily be applied to existing structured overlays unlike the concept of innermost and outermost circles. Our approach also enables the profile owner to serve as a CA, ensuring that nodes can only access a profile overlay after having obtained a signed certificate.

Unlike the above approaches, the P2P social network presented in [1] uses an unstructured overlay without a DHT where peers connect directly to each other rather than through the overlay establishing unique identifiers to deal with dynamic IPs. Peers cache each other’s data to improve availability. While helper nodes are used to assist with communication between peers behind NATs. The approach lacks security and access control considerations and lacks the guarantees and the simplicity of the abstraction offered by a structured overlay.

3 Social Overlays

In this section, we explain how to map online social networking to our multi-overlay social network consisting of a public directory overlay with many private profile overlays. The directory overlay supports friend discovery and verification and stores a lists of peers currently active in each profile overlay. Profile overlays support message boards, private messages, and media sharing.

3.1 Finding and Verifying Friends

In a traditional social network, a directory provides the ability to search for users using public information, such as the user’s full name, user ID, e-mail address, group affiliations, and friends. The search results return zero or more matching directory entries. Based upon the results, the user, A, can potentially make a friendship request. The request receiver, B, can review the public information of A to making a decision. If B accepts the request, A and B are given access to each other’s profiles. Once profile access has been enabled, the users can learn more information, and if it turns out to be a mistake, the peers can unilaterally end the relationship.

To map this to our proposed social overlay, the directory entries would be inserted into the DHT of a public overlay. As discussed in previous work, the DHT keys for these entries should consist of a subset of the user’s public information in lower-case format and hashed to an overlay address. The value stored at these keys is the user’s certificate, which consists of its public information and an overlay address where the user expects to receive notifications. This overlay address can be used for asynchronous offline messaging, whose function we will explain shortly.

Because the users need a way to verify each other that involves social credentials, we propose the use of a new form of certificate. The main portion of the certificate is similar to a self-signed x509 certificate with public information such as user’s name, e-mail address, and group affiliations embedded into the certificate. At the tail of the certificate is a friend list represented by many friend entries. To do this we propose employing a technique similar to PGP: users can acquire from their friends a signed message consisting of a hash of the peer’s certificate, the time stamp, and the friend’s certificate hash signed by the friend. Since PGP does not provide a strong method for revocation, the time stamp provides additional information to help decide whether or not a friendship link is still active without accessing the profile overlay of either peers. Peers should request a new friend list entry within a certain period of time or it will appear that the friendship is no longer valid.

While looking for an individual, a peer may discover that many individuals have overlapping public information components, such as the user’s name. Assuming all entries are legitimate, the overlay must have some method of supporting multiple, distinct values at the same key, leaving the peer or the peer’s DHT client to parse the responses and determining the best match by reviewing the contents of each certificate. Alternatively, a technique like Sword [2], which supports distributing the data across a set of nodes and using a bounded broadcast to discover peers that match all information, could be used for searching.

If a peer, A, desires a friendship with another peer, B, A issues a friendship request, which will be stored in the DHT at the overlay address listed in B’s certificate, as described earlier. The friendship request consists of the self-signed certificate of A, the requesting peer; the public information of the request receiver, B; and a time stamp; all signed with the private key associated with A’s private key matched to their self-signed certificate.

Within a reasonable amount of time after a request has been inserted into the DHT, B can come online and check for outstanding requests. Upon receiving a request, B has three choices: a conditional accept, an unconditional accept, or a reject. During an unconditional accept, B signs A’s request and issues a request to befriend A. Alternatively in the case of a conditional accept, B issues a friendship request, waits for a reply, and investigates the profile prior to signing the A’s request. Once a user has received a signed certificate, they may access the remote peer’s profile overlay as discussed in 3.2, which is also responsible for activities such as revocation.

3.2 The Profile Overlay

In a traditional social network, the profile or user-centric portion consists of private messaging, data sharing, friendship maintenance, and a public message board for status updates or public messages. In this section, we explain how these components can be applied to a structured overlay dedicated to an individual profile.

Using the techniques such as those described in [22], it is feasible to efficiently multiplex a P2P system across multiple, virtual private overlays enabling each profile owner to have a profile overlay consisting of their online friends. For access control, we employ a PKI, where each member uses the signed certificate generated during the “finding and verifying friends” stage. All links are encrypted using symmetric security algorithms established through the PKI, thus preventing uninvited guests from gaining direct access to the overlay and hence the profile. Because the profile owner also is the CA for all members of the overlay, they can easily revoke users from access to the profile overlay. In  [22], we described mechanisms for overlay revocation through the use of broadcasting for immediate revocation and the use of DHT for indirect and permanent revocation.

The message board of a profile can be stored in two ways: distributed within the profile overlay via a data store or stored on the profile owner’s personal computing devices. The distributed data store provide the profile when the owner is offline and also distributes the load for popular profiles. For higher availability, each peer should always be a provider for all data in their profile when they are online. To ensure authenticity and integrity, all peers should sign their messages and each peer’s certificate should be available in the overlay for verification. Messages that are unsigned should be ignored by all members of the overlay. An ideal overlay for this purpose should support complex queries [11] allowing easy access to data stored chronologically, by content, by type, i.e., media, status updates, or message board discussions.

Private messaging in the profile overlay is unidirectional meaning that only the profile owner can receive private messages using their overlay. To enforce this, a private message should be prepended with a symmetric key encrypted by the profile owners public key, the message should be appended by a hash of the message to ensure integrity and the entire message encrypted by the symmetric key. This approach ensures that only the sender and the profile owner can decrypt the private message. The contents of the private message should include the sender, time sent, and the subject. Messages can be stored in well known locations, so that the profile owner can either poll the location or, alternatively, use an event based system to notify them of the new message.

3.3 Active Peers

The directory overlay should be used to assist in finding currently active peers in the profile overlays. By placing their node IDs at a well-known, unique per-profile overlay keys in the DHT, active peers can bootstrap incoming peers into the profile overlay. We implemented and evaluated this concept in [22]. Because the profile overlay members all use PKI to ensure membership, even if malicious peers insert their ID into the active list, it would be useless as the peer would only form connections with peers who also have a signed certificate.

4 Challenges

While structured P2P overlays have been well-studied in a variety of applications, their use in social profile overlays raises new interesting questions, including:

1) Handling small overlay networks - P2P overlay research typically focuses on networks larger than the typical user’s friend count (Facebook’s average is 130111 Because social profile overlays are comparatively smaller, this can impact the reliability of the overlay and availability of profile data. A user can host their own profile; however when the user is disconnected it is important that their profile remains available even under churn. It is thus important to characterize churn in this application to understand how to best approach this problem. An optional of per-user deployment of a virtual individual server (VIS) and the use of replication schemes aware of a user’s resources provide possible directions to address this issue.

2) Overlay support for low bandwidth, unconnected devices - devices such as smart phones cannot constantly be actively connected to the overlay and the connection time necessary to retrieve something like a phone number may be too much to make this approach useful. Similar to the previous challenge, this approach could benefit from using a VIS enabling users access to their social overlays by proxy without establishing a direct connection to the overlay network.

3) Reliability of the directory and profile overlay - Overlays are susceptible to attacks that can nullify their usefulness. While the profile overlay does have point-to-point security, in the public, directory overlay, the lack of any form centralization makes policing the system a complicated procedure. While our approach of appending friends list can assist users in making decisions on identity, it does not protect against denial of service attacks. For example, users could attempt create many similar identities in an attempt to overwhelm a user in their attempt to find a specific peer. Previous work has proposed methods to ensure the usability of overlays even while under attack. For the social overlay to be successful, we must identify which methods should be used. A possible approach is to replicate public information within a user’s profile overlay thus providing an alternative directory overlay for querying prior to using the public directory overlay.

5 Conclusion

In this paper, we proposed methods by which a social network can be decentralized through the use of structure P2P overlays. Our approach uses multiple overlays where all users join a public directory overlay and a subset of the individual profile overlays. The directory overlay enables users to find and befriend other peers and bootstrap connections into the secure profile overlays. Upon forming a friendship through the directory overlay, peers are given CA signed certificates that allow them to join each other’s profile overlay. The owner of the profile overlay acts as CA enabling unilateral dismissal of friendships via certificate revocation using efficient and reliable methods. For the purpose of storing profile information into the overlay, we cite previous work that can be used to provide distributed data services and give examples of how to store data securely in the overlay. Our proposed system returns control of the social network and more importantly users’ identity to the users and eliminates the need for centralized social networks.


  • Abbas et al. [2009] S. M. A. Abbas, J. A. Pouwelse, D. H. J. Epema, and H. J. Sips. A gossip-based distributed social networking system. In Enabling Technologies, IEEE International Workshops on, 2009.
  • Albrecht et al. [2008] J. Albrecht, D. Oppenheimer, A. Vahdat, and D. A. Patterson. Design and implementation trade-offs for wide-area resource discovery. In ACM Trans. Internet Technol., 2008.
  • Buchegger and Datta [2009] S. Buchegger and A. Datta. A case for P2P infrastructure for social networks - opportunities & challenges. In WONS ’09: The Sixth International Conference on Wireless On-demand Network Systems and Services, 2009.
  • Buchegger et al. [2009] S. Buchegger, D. Schiöberg, L. H. Vu, and A. Datta. Peerson: P2p social networking: early experiences and insights. In SNS ’09: Proceedings of the Second ACM EuroSys Workshop on Social Network Systems, 2009.
  • [5] A. R. Butt, T. A. Johnson, Y. Zheng, and Y. C. Hu. Kosha: A peer-to-peer enhancement for the network file system. In IEEE/ACM Supercomputing 2004.
  • [6] M. Castro, M. Costa, and A. Rowstron. Debunking some myths about structured and unstructured overlays. In NSDI’05: Proceedings of Symposium on Networked Systems Design & Implementation.
  • Castro et al. [2002a] M. Castro, P. Druschel, A. Ganesh, A. Rowstron, and D. S. Wallach. Security for structured peer-to-peer overlay networks. In Symposium on Operating Systems Design and Implementaion (OSDI’02), December 2002a.
  • Castro et al. [2002b] M. Castro, P. Druschel, A.-M. Kermarrec, and A. Rowstron. One ring to rule them all: Service discover and binding in structured peer-to-peer overlay networks. In SIGOPS European Workshop, Sept. 2002b.
  • [9] L. A. Cutillo, R. Molva, and T. Strufe. Privacy preserving social networking through decentralization. In Wireless On-Demand Network Systems and Services (WONS’09).
  • DeCandia et al. [2007] G. DeCandia, D. Hastorun, M. Jampani, G. Kakulapati, A. Lakshman, A. Pilchin, S. Sivasubramanian, P. Vosshall, and W. Vogels. Dynamo: amazon’s highly available key-value store. In SOSP ’07: Proceedings of twenty-first ACM SIGOPS symposium on Operating systems principles, pages 205–220, New York, NY, USA, 2007. ACM. ISBN 978-1-59593-591-5. doi:
  • Harren et al. [2002] M. Harren, J. M. Hellerstein, R. Huebsch, B. T. Loo, S. Shenker, and I. Stoica. Complex queries in dht-based peer-to-peer networks. In IPTPS ’01: Revised Papers from the First International Workshop on Peer-to-Peer Systems, 2002.
  • Manku et al. [2003] G. S. Manku, M. Bawa, and P. Raghavan. Symphony: distributed hashing in a small world. In USITS, 2003.
  • Maymounkov and Mazières [2002] P. Maymounkov and D. Mazières. Kademlia: A peer-to-peer information system based on the XOR metric. In IPTPS ’02, 2002.
  • Perez [2007] J. C. Perez. Facebook’s beacon more intrusve than previously thought., 2007.
  • [15] S. Ratnasamy, M. Handley, R. M. Karp, and S. Shenker. Application-level multicast using content-addressable networks. In Workshop on Networked Group Communication (NGC’01).
  • Ratnasamy et al. [2001] S. Ratnasamy, P. Francis, S. Shenker, and M. Handley. A scalable content-addressable network. In In Proceedings of ACM SIGCOMM, 2001.
  • [17] A. Rowstron and P. Druschel. Storage management and caching in PAST, a large-scale, persistent peer-to-peer storage utility. In Symposium on Operating Systems Principles (SOSP’01).
  • Rowstron and Druschel [2001] A. Rowstron and P. Druschel. Pastry: Scalable, decentralized object location and routing for large-scale peer-to-peer systems. In IFIP/ACM International Conference on Distributed Systems Platforms (Middleware), November 2001.
  • Shakimov et al. [2008] A. Shakimov, H. Lim, L. P. Cox, and R. Caceres. Vis-à-vis:online social networking via virtual individual servers. Technical report, May 2008.
  • Stoica and et al. [2001] I. Stoica and et al. Chord: A scalable Peer-To-Peer lookup service for internet applications. In SIGCOMM, 2001.
  • Wolinsky et al. [2009a] D. I. Wolinsky, L. Abraham, K. Lee, Y. Liu, J. Xu, P. O. Boykin, and R. Figueiredo. On the design and implementation of structured P2P VPNs. In TR-ACIS-09-003, October 2009a.
  • Wolinsky et al. [2009b] D. I. Wolinsky, K. Lee, T. W. Choi, P. O. Boykin, and R. Figueiredo. Virtual private overlays: Secure group communication in NAT-constrained environments. In TR-ACIS-09-004, December 2009b.
Comments 0
Request Comment
You are adding the first comment!
How to quickly get a good reply:
  • Give credit where it’s due by listing out the positive aspects of a paper before getting into which changes should be made.
  • Be specific in your critique, and provide supporting evidence with appropriate references to substantiate general statements.
  • Your comment should inspire ideas to flow and help the author improves the paper.

The better we are at sharing our knowledge with each other, the faster we move forward.
The feedback must be of minimum 40 characters and the title a minimum of 5 characters
Add comment
Loading ...
This is a comment super asjknd jkasnjk adsnkj
The feedback must be of minumum 40 characters
The feedback must be of minumum 40 characters

You are asking your first question!
How to quickly get a good answer:
  • Keep your question short and to the point
  • Check for grammar or spelling errors.
  • Phrase it like a question
Test description