Presentation is loading. Please wait.

Presentation is loading. Please wait.

Traffic Morphing: An Efficient Defense Against Statistical Traffic Analysis Presented by Yang Gao 11/2/2011 Charles V. Wright MIT Lincoln Laboratory Scott.

Similar presentations


Presentation on theme: "Traffic Morphing: An Efficient Defense Against Statistical Traffic Analysis Presented by Yang Gao 11/2/2011 Charles V. Wright MIT Lincoln Laboratory Scott."— Presentation transcript:

1 Traffic Morphing: An Efficient Defense Against Statistical Traffic Analysis Presented by Yang Gao 11/2/2011 Charles V. Wright MIT Lincoln Laboratory Scott E. Coull Johns Hopkins University Fabian Monrose University of North Carolina

2 Outline  Potential Hazards  Counter measures and Traffic Morphing  How it works?  Evaluation and Results

3 Privacy Security

4 Packet Size and Timing Information Privacy Leakage Classification Tools Language of a VoIP call Password in SSH Web browsing habits...

5 How does the attack happen  Webpage browsing  Statistical Identification of Encrypted Web Browsing Traffic (Sun,Q. Stanford University)

6 A 2000 sample from 100,000 WebPages Only Objects number and sizes are recorded Jaccard’s coefficient Trained classifier

7 How does the attack happen  Webpage browsing  Statistical Identification of Encrypted Web Browsing Traffic (Sun,Q. Et Stanford University)  Inferring the Source of Encrypted HTTP Connections (Marc Liberatore and Brian Neil Levine UMA)  Identification of Encrypted VoIP Traffic

8 Results of the Classifiers

9 Outline  Potential Hazards  Counter measures and Traffic Morphing  How it works?  Evaluation and Results

10 Countermeasures  Padding  Mimicking  Morphing  Sending at fixed time intervals(counter the timing analysis)

11 Comparison

12 Traffic Morphing morphing

13 How does the morphing work? L1L2L1 L2L1L2 N L1 : N L2 = 2 : 1 N L1 : N L2 = 1 : 2

14 Outline  Potential Hazards  Counter measures and Traffic Morphing  How it works?  Evaluation and Results

15 Traffic Morphing  Goals  Good resemblance in packet size distribution  Less overhead  Steps  Morphing matrix construction

16 Morphing Matrix Size x1 Size xn Size y1 Size yn 2*n equations and n 2 unknowns

17 How to solve these equations?  We won't solve them directly.  Convex Optimization  Cost Function  Restrictions

18 Example L1L2L1 L2L1L2

19 Example L1L2L1 L2 Reduce? Add more constrains to avoid this situation.

20 Steps for Traffic Morphing  Matrix Construction  Select the source process and calculate the probability distribution of the packets size.  Select the target process and calculate the probability distribution of the packets size.  Solve the morphing matrix with optimization method which could minimize the cost while following the restrictions.  Traffic Morphing  Get the packet to send.  set up a random number to select the element in the matrix  Calculate the corresponding packet size.  Padding or reduce the packet size  Transmit the new packet.

21 Traffic Morphing  Goals  Good resemblance in packet size distribution  Less overhead  Steps  Morphing matrix construction  Additional Morphing Constraints

22 Pitfall 1  System is over-specified  Y = AX  Solution:  Multi-level programming  Find Z which is closest to Y  Find A which such that most efficiently maps X to Z  Z=A’X => Minimize( f d (Y,Z) )  Z=AX => Minimize( f 0 (A) )

23 Traffic Morphing  Goals  Good resemblance in packet size distribution  Less overhead  Steps  Morphing matrix construction  Additional Morphing Constraints  Dealing with Large Sample Spaces

24 Pitfall 2  Pool Scalability  Pentium 4 2.8G run 1 hr for 80x80 matrix with 6560 constraints  MTU(40~1500) means 1460x1460 Matrix  Solution  Multi-level method  Sub-matrix Morphing

25 Multi-level method

26 Traffic Morphing in sum  Goals  Good resemblance in packet size distribution  Less overhead  Steps  Morphing matrix construction  Convex optimization  Additional Morphing Constraints  2 level Multi-level programming  Dealing with Large Sample Spaces  Sub-matrix Morphing

27 Outline  Potential Hazards  Counter measures and Traffic Morphing  How it works?  Evaluation and Results

28 Evaluation  Encrypted Voice over IP  Web Page Identification  Defeating Original Classifier  Evaluating Indistinguishability

29 Encrypted Voice over IP  Language Identification of Encrypted VoIP Traffic:Alejandra y Roberto or Alice and Bob?  Charles V. Wright Lucas Ballard Fabian Monrose Gerald M. Masson from Department of Computer Science Johns Hopkins University

30 White box encode

31 Why even the encrypted voice packet will leak information  Unigram frequencies of bit rates

32 2-gram resemblance

33 Blackbox

34 Results for original classifier

35 Results for Indistinguishablity

36 Overhead

37 Web page Identification

38 Overhead

39 Practical Considerations  Short Network Sessions  Short of packets generated by source?  Keep generating until reach a distance threshold  Variations in Source Distribution  Packets size difference for training and using?  Divide and conquer  Reduced Packet Sizes  How to deal with the reduced packet size in HTTP  Packing to the next

40 Traffic Morphing in a nut shell  Resemblance  Morphing Matrix  Convex Optimization  Overhead Minimization  Additional Morphing Constraints  Dealing with Large Sample Spaces  Practical Considerations  Short Network Sessions  Variations in Source Distribution  Reduced Packet Sizes

41 Conclusion  User privacy are vulnerable even under encryption protected.  Traffic morphing is effective and robust  Traffic morphing is applicable.  Traffic morphing is much more efficient than padding.

42 Discussion  The other side of morphing  Anti-intrude-detection.  Mimicry attack System call sequence Malicious call combination library deny accept morphing

43 Thank you!


Download ppt "Traffic Morphing: An Efficient Defense Against Statistical Traffic Analysis Presented by Yang Gao 11/2/2011 Charles V. Wright MIT Lincoln Laboratory Scott."

Similar presentations


Ads by Google