1. OblivP2P: An Oblivious Peer-to-Peer Content Sharing System
Yaoqi Jia, Tarik Moataz, Shruti Tople and Prateek Saxena
National University of Singapore

2. Traffic Analysis in P2P Systems
P2P content sharing systems
150 million users/month
3.35% of all world bandwidth
Long term global traffic analysis
E.g., ISP’s, Global BitTorrent Monitor, Bitstalker
In this work, we look at the area of trafffic analysis in P2P systems
P2P contenst sharing systems like BitTorrent and others have millions of users accounting to 3% of the world bandwidth
In these systems, users communicate with each other to share resources for easy access and availability reasons
However, these systems are susceptible to long term global traffic analysis by adversarial organizations coz they have access to network traffic or through global analysis tools
Further, some of the peers can also be adversarial and collude in the network.

3. What can an Adversary do?
Leakage Channels
Plaintext data
Secure channel
Length
Padding
Time
Fixed Interval
Access Patterns
Assume existing
defenses
Given such long term analysis capability, we ask what can an adversary do by observing communication pattern between peers?
The adversary can learn private information about the user through several direct and indirect channels like plaintext data, length of message and the time interval
Previous work has focussed on these channel and proposed solutions like using a secure channel, padding messages and fixed interval messaging.
Assuming that these defenses are already deployed in P2P systems, the adversary can still learn substantial information by observing the access patterns of users.
i.e., which peer is accessing what resource from which other peer in the network.
Hence a global adversary links sender to their receiver thereby learning information about their online behaviour.
Linkability

4. Problem Is anonymizing enough? Current Solutions
Anonymous Systems e.g., Mix Networks, Tor
Hide Online Identity
Unlinkability
Currently, anonymous systems such as mix networks, tor are used as defenses in P2P systems.
However, these systems are essentially designed to hide online idenities and fail to guarantee unlinkability in presence of a long term and global adversary, as I will demonstarte later
These forces us to ask the question : Is anonymizing enough?
Long term
Global
Adversary
Is anonymizing enough?

5. Contributions Evaluation OblivP2P Protocol
Guarantee unlinkability
Obliviousness in P2P systems
OblivP2P
Protocol
Link:
Implementation
As our contributions in this work, we present OblivP2P protocol that guarantees unlinkability by ensuring obliviousness in P2P systems.
We implement our protocol and have made it available online
Finally, we evaluate our solution and observe that the performance of OblivP2P has no centralized bottleneck and linearly scales with the number of peers
No Centralized Bottleneck
Linear Scalability with peers
Evaluation

6. Problem
Let us investigate the problem we are solving in depth

7. Insufficiency of Existing Solutions
Mixnet
Round 1
Mixnet
Round 2
Let us assume the use of mix network in a P2P network where the sender submit their resources to this mixnet. The mixnet shuffles and re-encrypts the data and distributes it to corresponding receivers.
A global adversary observing
Intersection, Hitting Set [AK’03] or
Statistical Disclosure Attacks [KP’04]

8. Main Insight: Oblivious Access Pattern
Oblivious RAM
Hide access patterns between CPU and memory
Data is shuffled in the memory periodically
Applied to:
Cloud Storage [SS’13A], [SS’13B],[LO’13]
Filesystem [WST’12]
Can we directly apply ORAM to P2P systems?

9. ORAM Background Tree-Based ORAM (Path ORAM) Read Write
Fetches a path from the tree containing the block
Stores the path in the local storage (stash)
Write
Selects a random path in the tree
Shuffles the blocks in the stash and the path 1 2 3 4 5 6 7

10. Mapping ORAM to P2P OblivP2P-0 Protocol Trusted Client Trusted Tracker
Re-encrypt
Send
Trusted Client
Trusted Tracker 2 2
Position Map, Stash
Position Map, Stash
Request
Fetch a path
Decrypt
Decrypt 2
Fetch path
Initiator
Untrusted Server 1
path 1 2 3 2 3 4 5
Peers 6 7 4 5 6 7
Peer-to-Peer Network
OblivP2P-0 Protocol

11. OblivP2P-0: Tracker as bottleneck
Tracker fetches O(log N) blocks per access
118 MB /req
for 2 million users
≈ Max Bandwidth
Few sentences
Need a Distributed Oblivious P2P Protocol

12. OblivP2P-1 Protocol

13. Naïve approach: Removing Bottleneck
Send
< path, position, key >
Trusted Tracker
Performance
Position Map, Stash
Security
Request
Decrypt 2
Initiator
Fetch a path 1 2 3 4 5
Peers 6 7
Peer-to-Peer Network

14. Less frequently accessed
Challenges
ORAM writes
Recently accessed block at the root
Less frequently accessed block at the leaves
“Block History”
Shared resources
Security flaw in P2P systems
Multiple users access the same resource
Recently accessed 1 2
Less frequently accessed 5

15. New Primitive: Oblivious Selection
Selects a block without :
Block Position
Cryptographic Key
No Centralized Bottleneck

16. Construction Step 1: PIR over ORAM
Obliviously select a block from a path
Trusted Tracker 5 1 2 5 1 2 5 1 2
Send PIR metadata
Request 2
Compute an Encrypted Share using PIR
Initiator
Block Position
Cryptographic Key
No Centralized Bottleneck

17. Construction Step 2: Seed-Homomorphic PRG
Decrypt shares without giving away the key
Trusted Tracker
Send key share
Compute a Decrypted Share using SH-PRG
Initiator
Dec Block
Block Position
Cryptographic Key
Cryptographic Key
No Centralized Bottleneck

18. Security

19. OblivP2P is an Oblivious P2P Protocol
Any two equal length access sequences by two peers are indistinguishable for any p.p.t. “honest-but-curious” adversary
Number of dishonest peers is in O(Nε), where ε<1
Theorem: If ∀N > 1, and ∀ε < 1, ∃m > 1 such that 2logN·m·(1−ε) ∈ negl(λ) then OBLIVP2P-1 is an oblivious P2P protocol

20. Evaluation

21. Experimental Setup 15 DeterLab servers – (214)16000 peers
Each server shares a bandwidth of 128 MBps
Block size of 512 KB similar to BitTorrent

22. No Centralized Bottleneck
~ 128 MB /req
~ 1 MB /req

23. Linear Scalability with Peers
3.59MB/ sec
Larger networks can scale up performance
3.59 MB/s is due to our limited test infrastructure
Bottleneck remaining is purely computational

24. Take Away! Propose hiding data access patterns in P2P systems
OblivP2P - First work to repurpose ORAM in Peer-to-Peer systems
OblivP2P is linearly scalable and highly parallelizable with the peers in the network

25. Link: https://github.com/jiayaoqijia/OblivP2P-Code
Thanks!
Link: