Apparatus method and medium for detecting payload anomaly using n-gram distribution of normal data

• US 7,639,714 B2
• Filed: 11/12/2004
• Issued: 12/29/2009
• Est. Priority Date: 11/12/2003
• Status: Active Grant

• Current Assignee: Trustees Of Columbia University In The City Of New York (Columbia University)
• Original Assignee: Trustees Of Columbia University In The City Of New York (Columbia University)
• Inventors: Stolfo, Salvatore J., Wang, Ke
• Primary Examiner(s): Moe, Aung S
• Assistant Examiner(s): Pasia, Redentor M
• Application Number:
  US10/986,447
• Publication Number:
  US 20050281291A1
• Time in Patent Office: 1,873 Days
• Field of Search: 370229-231, 370/235, 370/250, 370252-253, 370/389, 370/392, 370/401, 370/476, 370470-472, 370/474, 713/176, 713188-189, 726/1, 726/3, 726/11, 726/13, 726/14, 726 22- 25, 709223-225
• US Class Current: 370/474
• CPC Class Codes: G06F 21/55 Detecting local intrusion or implementing counter-measures
• CPC Class Codes: G06F 21/552 involving long-term monitoring or reporting
• CPC Class Codes: G06F 21/554 involving event detection and direct action
• CPC Class Codes: G06F 21/56 Computer malware detection or handling, e.g. anti-virus arrangements
• CPC Class Codes: G06F 21/562 Static detection
• CPC Class Codes: G06F 21/563 by source code analysis
• CPC Class Codes: G06F 21/564 by virus signature recognition
• CPC Class Codes: H04L 43/00 Arrangements for monitoring or testing data switching networks
• CPC Class Codes: H04L 43/0876 Network utilisation, e.g. volume of load or congestion level
• CPC Class Codes: H04L 63/0218 Distributed architectures, e.g. distributed firewalls
• CPC Class Codes: H04L 63/0245 Filtering by information in the payload
• CPC Class Codes: H04L 63/0263 Rule management
• CPC Class Codes: H04L 63/029 Firewall traversal, e.g. tunnelling or, creating pinholes
• CPC Class Codes: H04L 63/1425 Traffic logging, e.g. anomaly detection
• CPC Class Codes: H04L 63/145 the attack involving the propagation of malware through the network, e.g. viruses, trojans or worms

Abstract

A method, apparatus and medium are provided for detecting anomalous payloads transmitted through a network. The system receives payloads within the network and determines a length for data contained in each payload. A statistical distribution is generated for data contained in each payload received within the network, and compared to a selected model distribution representative of normal payloads transmitted through the network. The model payload can be selected such that it has a predetermined length range that encompasses the length for data contained in the received payload. Anomalous payloads are then identified based on differences detected between the statistical distribution of received payloads and the model distribution. The system can also provide for automatic training and incremental updating of models.

Claims

1. A method, implemented on a processor, of detecting anomalous payloads transmitted through a network, comprising the steps of:

receiving at least one payload within the network;

determining, using the processor, a length for data contained in the at least one payload;

generating, using the processor, a byte value statistical distribution of data contained in the at least one payload received within the network;

selecting, using the processor, from a plurality of model byte value statistical distributions, a model byte value statistical distribution representative of normal payloads transmitted through the network based at least in part on the determined length, wherein the model byte value statistical distribution has a predetermined length range and is selected such that the determined length for the data contained in the at least one payload is included within the predetermined length range;

comparing, using the processor, at least one portion of the generated statistical distribution to a corresponding portion of a selected model distribution; and

identifying, using the processor, whether the at least one payload is an anomalous payload based, at least in part, on differences detected between the at least one portion of the byte value statistical distribution for the at least one payload and the corresponding portion of the selected model distribution.

2. The method of claim 1, wherein the predetermined length range of the selected model distribution is based on partitions selected using kernel estimates for the byte value statistical distribution of all normal payloads.

3. The method of claim 1, wherein the predetermined length range of the selected model distribution is based on partitions selected by applying at least one clustering algorithm to the byte value statistical distribution of all normal payloads.

4. The method of claim 1, wherein byte value the statistical distribution of the at least one payload is a byte value distribution of the average frequency and variance of data contained in the at least one payload, and the selected model distribution is a byte value distribution of the average frequency and variance representative of normal payloads.

5. The method of claim 1, wherein the byte value statistical distribution of the at least one payload is a byte value distribution of data contained in the at least one payload, and the selected model distribution is a byte value distribution representative of normal payloads.

6. The method of claim 5, wherein the byte value distribution of the at least one payload is a byte frequency count of data contained in the at least one payload, and the selected model distribution is a byte frequency count for normal payloads.

7. The method of claim 5, wherein the byte value distribution of the at least one payload is a rank ordered byte frequency count of data contained in the at least one payload, and the selected model distribution is a rank ordered byte frequency count for normal payloads.

8. The method of claim 1, wherein:

the step of comparing further comprises a step of measuring a distance metric between the byte value statistical distribution of data contained in the at least one payload and the selected model distribution; and

the step of identifying further comprises a step of identifying anomalous payloads as payloads that, at least in part, exceed a predetermined distance metric.

9. The method of claim 8, wherein the distance metric is calculated based on a Mahalanobis distance between the byte value statistical distribution of data contained in the at least one payload and the selected model distribution.

10. The method of claim 8, wherein the byte value statistical distribution of data contained in the at least one payload has a decaying profile, and wherein measurement of the distance metric is initiated at the decaying end.

11. The method of claim 8, further comprising the steps of:

setting an alert threshold corresponding to a desired number of alerts for detected anomalous payloads; and

automatically adjusting the predetermined distance metric until the alert threshold is reached.

12. The method of claim 1, wherein the predetermined length range is selected from at least a prefix portion of the selected model distribution.

13. The method of claim 1, wherein the predetermined length range is selected from at least a suffix portion of the selected model distribution.

14. The method of claim 1, wherein the predetermined length range is selected from at least a central portion of the selected model distribution.

15. The method of claim 1, wherein the predetermined length range of the selected model distribution is selected from a partition group consisting of:

0-50 bytes, 50-150 bytes, 150-155 bytes, 0-255 bytes, less than 1000 bytes, greater than 2000 bytes, and greater than 10,000 bytes.

16. The method of claim 1, wherein the byte value statistical distribution for the at least one payload and a statistical distribution of the selected model distribution are generated using an n-gram distribution, wherein each n-gram is a variable byte grouping.

17. The method of claim 16, wherein n is a mixed value of byte groupings.

18. The method of claim 16, wherein n=1.

19. The method of claim 16, wherein n=2.

20. The method of claim 16, wherein n=3.

21. The method of claim 1, further comprising a step of determining whether a detected anomalous payload is a virus or worm.

22. The method of claim 21, further comprising the steps of:

identifying a port which has been probed by the virus or worm;

comparing egress traffic to the byte value statistical distribution for the virus or worm in order to detect probes to the same port on other machines; and

detecting propagation of the worm or virus based on the step of comparing.

23. The method of claim 21, further comprising a step of assigning different weight factors to selected byte values.

24. The method of claim 23, wherein higher weight factors are assigned to byte values corresponding to operational codes of a computer system.

25. The method of claim 21, further comprising a step of generating a signature for any anomalous payload determined to be the virus or worm.

26. The method of claim 25, wherein the step of generating a signature further comprises the steps of:

identifying the longest common string for data contained in the payload determined to be the virus or worm; and

generating the signature based, at least in part, on the longest common string.

27. The method of claim 25, wherein the step of generating further comprises the steps of:

identifying the longest common subsequence for data contained in the payload determined to be the virus or worm; and

generating the signature based, at least in part, on the longest common subsequence.

28. The method of claim 25, further comprising a step of exchanging virus or worm signatures with one or more servers.

29. A system for detecting anomalous payloads transmitted through a network, comprising:

a computer coupled to the network, and receiving at least one payload through the network; and

one or more model distributions representative of normal payloads received through the network;

said computer being configured to;

determine a length for data contained in said at least one payload,

generate a byte value statistical distribution of data contained in said at least one payload received within the network,

selecting, from the one or more model distributions, a model byte value statistical distribution based at least in part on the determined length, wherein the model byte value statistical distribution has a predetermined length range and is selected such that the determined length for the data contained in the at least one payload is included in the predetermined length range;

compare at least one portion of said generated byte value statistical distribution to a corresponding portion of said selected model byte value statistical distribution of said one or more model distributions, and

identify whether said at least one payload is an anomalous payload based, at least in part, on differences detected between the at least one portion of the byte value statistical distribution for said at least one payload and the corresponding portion of said selected byte value model distribution.

30. The system of claim 29, wherein said computer is further configured to:

measure a distance metric between the byte value statistical distribution of data contained in said at least one payload and said selected byte value model distribution; and

identify anomalous payloads as payloads that, at least in part, exceed a predetermined distance metric.

31. The system of claim 29, wherein said computer is further configured to:

set an alert threshold corresponding to a desired number of alerts for detected anomalous payloads; and

automatically adjust said predetermined distance metric until said alert threshold is reached.

32. The system of claim 29, wherein said computer comprises at least one port and is further configured to:

determine whether a detected anomalous payload is a virus or worm;

identify one of said ports which has been probed by said virus or worm; and

detect propagation of said worm or virus by comparing egress traffic to the byte value statistical distribution for said virus or worm in order to detect probes to the same port on at least one other computer.

33. A computer readable storage medium storing instructions executable by a computer for modeling payload data received in a network, the instructions causing said computer to perform the method of:

receiving at least one payload through the network;

determining a length for data contained in the at least one payload;

generating a byte value statistical distribution of data contained in the at least one payload received within the network;

selecting, from a plurality of model byte value statistical distributions, a model byte value statistical distribution representative of normal payloads transmitted through the network based at least in part on the determined length, wherein the model byte value statistical distribution has a predetermined length range and is selected such that the determined length for the data contained in the at least one payload is included within the predetermined length range;

comparing at least one portion of the generated statistical distribution to a corresponding portion of the selected model distribution; and

identifying whether the at least one payload is an anomalous payload based, at least in part, on differences detected between the at least one portion of the byte value statistical distribution for said the at least one payload and the corresponding portion of the selected model distribution.

34. The computer readable storage medium of claim 33, further comprising instructions for causing said computer to perform the acts of:

measuring a distance metric between the byte value statistical distribution of data contained in said at least one payload and said selected model distribution; and

identifying anomalous payloads as payloads that, at least in part, exceed a predetermined distance metric.

35. The computer readable storage medium of claim 33, further comprising instructions for causing said computer to perform the acts of:

setting an alert threshold corresponding to a desired number of alerts for detected anomalous payloads; and

automatically adjusting said predetermined distance metric until said alert threshold is reached.

36. The computer readable storage medium of claim 33, further comprising instructions for causing said computer to perform the acts of:

determining whether a detected anomalous payload is a virus or worm;

identifying one of said ports which has been probed by said virus or worm; and

detecting propagation of said worm or virus by comparing egress traffic to the byte value statistical distribution for said virus or worm in order to detect probes to the same port on at least one other computer.

37. A system for detecting anomalous payloads transmitted through a network, comprising:

means for receiving at least one payload through the network;

means for determining a length for data contained in said at least one payload;

means for generating a byte value statistical distribution of data contained in said at least one payload received within the network;

means for selecting, from a plurality of model byte value statistical distributions, a model byte value statistical distribution representative of normal payloads transmitted through the network based at least in part on the determined length, wherein the model byte value statistical distribution has a predetermined length range and is selected such that the determined length for the data contained in the at least one payload is included within the predetermined length range;

means for comparing at least one portion of the generated statistical distribution to a corresponding portion of the selected model distribution; and

means for identifying whether the at least one payload is an anomalous payload based, at least in part, on differences detected between the at least one portion of the byte value statistical distribution for said the at least one payload and the corresponding portion of said the selected model distribution.

38. The system of claim 37, further comprising:

means for measuring a distance metric between the byte value statistical distribution of data contained in said at least one payload and said selected model distribution; and

means for identifying anomalous payloads as payloads that, at least in part, exceed a predetermined distance metric.

39. The system of claim 37, further comprising:

means for setting an alert threshold corresponding to a desired number of alerts for detected anomalous payloads; and

means for automatically adjusting said predetermined distance metric until said alert threshold is reached.

40. The system of claim 37, further comprising:

means for determining whether a detected anomalous payload is a virus or worm;

means for identifying one of said ports which has been probed by said virus or worm; and

means for detecting propagation of said worm or virus by comparing egress traffic to the byte value statistical distribution for said virus or worm in order to detect probes to the same port on at least one other computer.