Efficient Content Verification in Named Data Networking

Transcription

1 Efficient Content Verification in Named Data Networking Dohyung Kim 1, Sunwook Nam 2, Jun Bi 3, Ikjun Yeom 1 mr.dhkim@gmail.com 1 Sungkyunkwan University 2 Korea Financial Telecommunications and Clearing Institute 3 singhua University

2 Named Data Networking (NDN) Name-based consumer-driven content delivery Request to Where Request for What Internet R1 R2

3 Named Data Networking (NDN) Name-based consumer-driven content delivery Request to Where Request for What Internet R1 R2 PIT entry is created at routers Interest

4 Named Data Networking (NDN) Name-based consumer-driven content delivery Request to Where Request for What Response R1 Internet R2 PIT-based Content Delivery In-network Caching

5 Named Data Networking (NDN) Name-based consumer-driven content delivery Request to Where Request for What Internet R2 Content is served from in-network cache R1 Interest

6 Secure Communication In IP networks R1 R2

7 Secure Communication In IP networks R1 R2 End-to-end secure channel

8 Secure Communication In IP networks R1 R2 End-to-end secure channel In NDN R1 R2

9 Secure Communication In IP networks R1 R2 End-to-end secure channel In NDN R1 R2 Content itself should be secure

10 Content Poisoning Attack Fabricated content is placed in the content store Router compromise

11 Content Poisoning Attack Fabricated content is placed in the content store Router compromise Injection from attackers server

12 Content Poisoning Attack Fabricated content is placed in the content store Router compromise Injection from attackers server

13 Content Poisoning Attack Distribution of the fabricated content Internet Poisoned content R2 R1

14 Content Poisoning Attack Distribution of the fabricated content Internet Poisoned content R2 R1 Interest

15 Content Poisoning Attack Distribution of the fabricated content Intenet Poisoned content is distributed by the system itself R2 R1

16 Content Poisoning Attack Distribution of the fabricated content Intenet Poisoned content is distributed by the system itself Users are separated from valid content sources R2 R1 Not forwarded Interest Poisoned response

17 NDN Content Verification Signature verification

18 NDN Content Verification Signature verification incurs huge computational overhead

19 Related Work Probabilistic caching - Bianchi, Giuseppe, et al. "Check before storing: What is the performance price of content integrity verification in LRU caching?." ACM SIGCOMM Computer Communication Review 43.3 (2013): Verification overhead is controlled by caching probability

20 Related Work Probabilistic caching - Bianchi, Giuseppe, et al. "Check before storing: What is the performance price of content integrity verification in LRU caching?." ACM SIGCOMM Computer Communication Review 43.3 (2013): Verification overhead is controlled by caching probability Limitation - Recency problem under dynamic content popularity - Limited application Strongly bounded with random caching policy

21 Motivations Why do we verify even the content that is not actually served?

22 Motivations Why do we verify even the content that is not actually served? ns-3 simulation for estimating the amount of serving contents Cache hit rate Proportion of serving content in the CS

23 Objective Reduce verification overhead while preserving functionality of the built-in signature verification

24 The Proposed Scheme Verify serving contents only

25 The Proposed Scheme - Verify Serving Contents Only

26 The Proposed Scheme - Verify Serving Contents Only

27 The Proposed Scheme - Verify Serving Contents Only Signature verification

28 The Proposed Scheme - Verify Serving Contents Only In the proposed scheme, poisoned content is either - Evicted from the content store without any damages to the network - Discarded by the verification mechanism before being brought out to the network

29 The Proposed Scheme Flag for the already verified content Content store structure verify False True name C1Name C2Name data

30 The Proposed Scheme Favor the already-verified content in the content store - Segmented LRU prevents serving content from being evicted by by-passing content in the content store

31 Efficiency Analysis Efficiency metric - : the number of examined poisoned contents - : the number of verifications

32 Efficiency Analysis In the basic scheme, corresponds to the proportion of the requests for the poisoned contents,

33 Efficiency Analysis In the basic scheme, corresponds to the proportion of the requests for the poisoned contents, In the proposed scheme, - is the request arriving rate - is the hit ratio for the unverified contents in the CS

34 Efficiency Analysis Hit ratio for the unverified contents Proportion of requests for content i

35 Efficiency Analysis Hit ratio for the unverified contents Cache-miss probability for the content i

36 Efficiency Analysis Hit ratio for the unverified contents Cache-hit probability for the content i

37 Efficiency Analysis Hit ratio for the unverified contents - According to Che approximation Cache-hit probability for the content i is the size of CS, and t is the residing time in the CS

38 Analytic Results without SLRU

39 Analytic Results with SLRU

40 Analytic Results In the proposed scheme without SLRU In the proposed scheme with SLRU When is close to 0, the proposed scheme achieve a 10 or 20 time larger value of The value of is changed according to the amount of poisoned content,

41 Evaluation Ns-3 simulation with Contents whose popularity follows Zipf-Mandelbrot distribution function - youtube trace from UMASS Campus during Mar in 2008

42 Results - Poisoned contents

43 Results - Effect of Segmented LRU

44 Results - youtube Trace

45 Results - youtube Trace

46 Discussion The access delay is increased due to the verification process - Limited to the first access to the content

47 Discussion The access delay is increased due to the verification process - Limited to the first access to the content Multiple pending interests may distribute the poisoned content - Verification for the content that is matched with multiple pending interests

48 Discussion The access delay is increased due to the verification process - Limited to the first access to the content Multiple pending interests may distribute the poisoned content - Verification for the content that is matched with multiple pending interests Cache is attacked by using unverified data - Might show abnormal cache-hit pattern, that is, different value of (Hit rate / amount of hit data)

49 Conclusion We look at the content poisoning attack in NDN Implementation and its effects

50 Conclusion We look at the content poisoning attack in NDN Implementation and its effects We present an efficient way to reduce overhead of content verification at routers Verification of the serving content only

51 Conclusion We look at the content poisoning attack in NDN Implementation and its effects We present an efficient way to reduce overhead of content verification at routers Verification of the serving content only We minimize verification overhead by favoring the serving contents in the CS Flag and Segmented LRU

52 Q and A Thank you