1. Cachet: A Decentralized Architecture for Privacy Preserving Social Networking with Caching Shirin Nilizadeh, 1 Sonia Jahid, 2 Prateek Mittal, 3 Nikita Borisov, 2 Apu Kapadia 1 1.Indiana University Bloomington 2.University of Illinois at Urbana-Champaign 3.University of California, Berkeley CoNEXT 2012

2. Online Social Networks (OSNs) have revolutionized the way we communicate 2 Credit: Dave Larson, Infographic: Spring 2012 Social Media User Statistics, http://blog.tweetsmarter.com/social-media/, May 2012

3. But at the cost of privacy: Users are not in control of their data 3

4. Objective A decentralized and efficient architecture for social networking that provides strong security and privacy guarantees 4

5. Security and Privacy Requirements 5 Confidentiality Content should be accessible to only those who are authorized Integrity No one should be able to perform unauthorized updates to the data Availability User content should remain available, even if the owner is offline, and despite potential malicious attempts to destroy the data Flexible Policies Fine grained access control, e.g.,“(friend AND co-worker) OR family” Relationship Privacy Relationships between users should remain hidden from third parties Adversary model Storage nodes untrusted with data & relationships, social contacts trusted DHTs can tolerate up to 25% of malicious nodes, focus is not on routing- based attacks, Sybil or DoS attacks

6. Existing mechanisms for decentralized privacy-preserving social networking 6  Without support for encryption -Diaspora [http://joindiaspora.com] -“…barring some revolutionary design as yet unthought-of…” -LotusNet [Aiello et al. ’12]  Lack fine-grained policies -SafeBook [Cutillo et al. ’09 ], PeerSoN [Buchegger et al. ’09]  Better policies, but high overhead of revocation, trusted stores -Persona [Baden et al. ’09]  Pure gossip, less availability -[Mega et al. 2011]

7. Basic Architecture 7 A slightly different version of this architecture was presented in 4 th IEEE International Workshop on Security and Social Networking (SESOC '12), Switzerland, Mar 19, 2012.

8. Users’ Information is stored in a Distributed Hash Table (DHT) 8 Bob Alice Carol Eve Alice’s Status update Visit Alice’s status Alice’s Status update

9. 9 Status Comment Video Link Cachet uses an object-oriented data structure ObjectDataPolicyReference List Status Message Object Like! Bob’s policy Alice’s Status Bob’s Comment Read Write/Delete Append

10. Attribute Based Encryption supports flexible confidentiality policies 10 friend, neighbor Bob CarolDiana AND Colleague Neighbor OR Friend colleague friend, colleague Alice  Attribute-based encryption (ABE) -User-defined attribute-based policy -We extend EASiER [Jahid et al. ‘11] (e.g., friend-of-friend attributes, thresholding)  Previous schemes do not provide flexible policies and efficient revocation

11. Providing confidentiality and integrity (read, write and append policies) 11 Alice’s status List of references to other objects Ref to Alice’s status = (objID; ABE(K; P); WPK) Ref to object n Alice’s wall … Sign with WSK Enc(K Alice, WSK) WPK ABE(ASK,AP) APK Symmetric Encrypted with K Ref to a comment Sign with ASK

12. Downloading and reconstructing a wall or an aggregated newsfeed is a lengthy process 1.Retrieve and ABDecrypt ‘update’ objects 2.Retrieve new objects from DHT 3.Decrypt updates with symmetric keys. Bob Alice Bob: Check my newsfeed Alice’s update: Ref. to Status 2000 Carol’s update: Ref. to Photo 54 Carol’s update: Ref. to Photo 54 Carol’s update: Ref. to Photo 54 Carol’s update: Ref. to Photo 54 ABDecryption Alice’s status: Status 2000 Alice’s status Status 2000 Carol’s photo 54 Carol’s photo: Photo 54 Symmetric Decryption Bob’newsfeed: Alice’s recent update Carol’s recent update. Bob’newsfeed: Alice’s recent update Carol’s recent update. 12

13. Social Caching leverage social trust relationships to reduce expensive decryption operations 13

14. Online social contacts can provide decrypted objects to contacts who also satisfy the policy 14 Alice’s Status Read Policy: Bob and Carol Alice’s Status Read Policy: Bob and Carol Alice’s Status Read Policy: Bob and Carol Alice’s Status Read Policy: Bob and Carol Alice’s Status Read Policy: Bob and Carol Alice’s Status Read Policy: Bob and Carol Bob Alice Carol Eve Alice’s Status Read Policy: Bob and Carol Alice’s Status Read Policy: Bob and Carol Alice’s Status Read Policy: Bob and Carol Alice’s Status Read Policy: Bob and Carol Alice’s Status Read Policy: Bob and Carol Alice’s Status Read Policy: Bob and Carol

15. Need to build the social overlay first: The presence protocol is also decentralized 15 Bob Alice Carol Eve Presence object: Presence object: Presence object

16. Gossip-based social caching algorithm: A greedy approach 1.Creating the Presence Table 2.Selecting a Contact; If all contacts are visited or known to be offline, proceed to step 7 3.DHT Lookup and Connection 4.Pulling Information 5.Caching Information 6.Updating Presence Table; returning to Step 2 to locate the next social contact to connect to. 1.Performing DHT Lookups for online social contacts with No Mutual Social Contacts 16 Bob Carol Eve Diana Mary Friend# mutual friends On/Off Bob25 Carol20 Mary13 Diana7 Eve2 Friend# mutual friends On/Off Bob25 On Carol20 Mary13 Diana7 Eve2 Friend# mutual friends On/Off Bob25 On Carol20 Off Mary13 Diana7 Off Eve2 On Presence object Update object Update object Alice Presence object Update object Update object Presence object Presence object Update object Update object Presence object Update object Update object Presence object Friend# mutual friends On/Off Bob25 On Carol20 Off Mary13 On Diana7 Off Eve2 On Update object Update object

17. We built a simulator to study cache performance Newsfeed application, prototype in Java Simulation environment – FreePastry Simulator – Social graph : Facebook friendship graph from the New Orleans regional network with 63,732 nodes and 1.54 million edges – 10%, 30%, 50% online friends Based on Skype statistics, the 10-30% range is more pertinent Performance metrics – hitRate = fraction of newsfeed constructed from cache – progressiveHitRate(d) = fraction of newsfeed after d lookups 17

18. Social caching provides most of the newsfeed… …but not all of it (need the DHT) 18

19. Most of the social cache’s benefit comes from the first ~15 DHT lookups 19 The Average Progressive Hit Rate for users who have 100 to 200 social contacts

20. Speedup of loading the newsfeed: 5X-10X (25X-50X for the first 80-90% updates) 20 Even with only 10% of social contacts online, social caching provides performance improvement.

21. Discussion: Can we deploy Cachet today? Privacy issues – Users are aware of being excluded, policies to linked objects are visible – Identities of users who satisfy a particular policy will be known to all of those identities Deployment challenges – node churn, NAT Replace Facebook? – Maybe, hopefully, but not necessarily the goal – Add {c|C}achet to P2P networks that use social links – Skype, Tor, BitTorrent, … 21

22. Cachet…  is a decentralized approach for privacy-preserving OSNs  uses attribute-based encryption for flexible confidentiality and integrity policies  is a DHT and a social-overlay hybrid for better availability 22 Acknowledgments and Disclaimers This material is based upon work supported by the National Science Foundation under Awards CNS- 0953655 and CNS-1115693, by the Boeing Trusted Software Center at the University of Illinois and by the National Security Agency. These sponsors don’t necessarily share our views. Shirin Nilizadeh Prateek MittalNikita BorisovSonia Jahid