Jordan University of Science and Technology Exscind: A...

37
Jordan University of Science and Technology 1 Monther Aldwairi and Duaa Alansari Exscind: A Faster Pattern Matching For Intrusion Detection Using Exclusion and Inclusion Filters 1 Seventh International Conference on Next Generation Web Services Practices (NWeSP’11) 19-21, October 2011 Salamanca, Spain Exscind @ NWeSP ‘11 by Monther Aldwairi and Duaa Alansari 10/14/2011

Transcript of Jordan University of Science and Technology Exscind: A...

Page 1: Jordan University of Science and Technology Exscind: A ...munzer/courses/INCS745/Lec/Exscind.pdfJordan University of Science and Technology 11 Monther Aldwairi and Duaa Alansari Exscind:

Jordan University of Science and Technology

11

Monther Aldwairi and Duaa Alansari

Exscind: A Faster Pattern Matching For Intrusion Detection Using Exclusion and

Inclusion Filters

1

Seventh International Conference on Next Generation Web Services

Practices (NWeSP’11)

19-21, October 2011

Salamanca, Spain

Exscind @ NWeSP ‘11 by Monther Aldwairi and Duaa Alansari 10/14/2011

Page 2: Jordan University of Science and Technology Exscind: A ...munzer/courses/INCS745/Lec/Exscind.pdfJordan University of Science and Technology 11 Monther Aldwairi and Duaa Alansari Exscind:

Jordan University of Science and TechnologyJordan University of Science and Technology

2

Introduction to IDS

� IDS inspects ingress/egress traffic for attack signatures/behaviour1. Drops the packet

2. Alerts the system security administrator

3. Logs malicious activity

� Anomaly-based IDS � Employs machine learning to profile, then classify traffic into either normal or abnormal behavior

++ Detects new attacks

- High false positives

- Longer detection time

� Signature-based IDS (Misuse)� Employs exact pattern matching to look for specific signatures within packets

+ Faster than anomaly but not enough for inline deployment

+ Low false positives

- But unable to detect new attacks

-Manual signatures generation

+- Accuracy depends on signatures accuracy

Exscind @ NWeSP ‘11 by Monther Aldwairi and Duaa Alansari 10/14/2011

Page 3: Jordan University of Science and Technology Exscind: A ...munzer/courses/INCS745/Lec/Exscind.pdfJordan University of Science and Technology 11 Monther Aldwairi and Duaa Alansari Exscind:

Jordan University of Science and TechnologyJordan University of Science and Technology

3

The Problem

� Snort rules or attack signatures

alert tcp $EXTERNAL_NET any -> $HTTP_SERVERS 80(msg:"WEB-IIS CodeRed v2 root.exe access";content:"/root.exe"; sid:1256)

Year 2003 [2] 2011

� 30%-60% of Snort processing time � pattern matching [1]

� Moore’s law! Internet speeds double every eighteen months

� Internet traffic is doubling every six months [4]

� Better Pattern Matching � Speed, Memory, Configurability

Exscind @ NWeSP ‘11 by Monther Aldwairi and Duaa Alansari 10/14/2011

Year 2003 [2] 2011

Rules 3000 11,416

Signatures 1542 9945

Page 4: Jordan University of Science and Technology Exscind: A ...munzer/courses/INCS745/Lec/Exscind.pdfJordan University of Science and Technology 11 Monther Aldwairi and Duaa Alansari Exscind:

Jordan University of Science and TechnologyJordan University of Science and Technology

Contributions

1. Add a modified BF in front of modified WM to skip as much clean traffic as possible

o Program ONLY the prefix (first n-gram) and store SIDs

o Produce a probable matches list

o Provide an index for the first match position

4

o Provide an index for the first match position

2. Modify the WM o Search ONLY the probable matches list

o Start from the first index of probable match

3. Support all ASCII codeso Extended, special characters and the Null character

4. Vary the prefix 4-to -10

Exscind @ NWeSP ‘11 by Monther Aldwairi and Duaa Alansari 10/14/2011

Page 5: Jordan University of Science and Technology Exscind: A ...munzer/courses/INCS745/Lec/Exscind.pdfJordan University of Science and Technology 11 Monther Aldwairi and Duaa Alansari Exscind:

Jordan University of Science and TechnologyJordan University of Science and Technology

Wu-Manber Algorithm

� Preprocessing Stageo Find the minimum pattern length (m)

o Choose Block size B (2 or 3)

o Builds three hash tables: SHIFT table, HASH table, and PREFIX table

o For any Block x, does it appear in any pattern?

5

x ∈∈∈∈ patterns? yes No

� q is the right most occurrence of x in a pattern

o HASH and PREFIX tables contain linked lists of signatures with zero shift

� Search Stageo Slide a window, w, over the packet ∈ SHIFT Table

o Hash last B chars of sliding window and access SHIFT table

� SHIFT[i]>zero slide the window and repeat

� If SHIFT[i]=zero, � search HASH and PREFIX tables to verify matches

Exscind @ NWeSP ‘11 by Monther Aldwairi and Duaa Alansari 10/14/2011

SHIFT [i] m-q m-B+1

Page 6: Jordan University of Science and Technology Exscind: A ...munzer/courses/INCS745/Lec/Exscind.pdfJordan University of Science and Technology 11 Monther Aldwairi and Duaa Alansari Exscind:

Jordan University of Science and TechnologyJordan University of Science and Technology

Wu-Manber Example6

Signature SID

"image/" 2706

"logged in" 162

"imagedata" 12280

"WINDIR" 3010

"SYSDIR" 3011

Block Shift Block Shiftmai 3 WIN 3mag 2 IND 2age 1 NDI 1ge/ 0 DIR 0log 3 SYS 3ogg 2 YSD 2gge 1 SDI 1

Exscind @ NWeSP ‘11 by Monther Aldwairi and Duaa Alansari 10/14/2011

Step z t i m a g e / l k S Y S D I R o shift output

1 z t i m a g 2

2 i m a g e / 0 image/

3 m a g e / l 4

4 / l k S Y S 3

5 S Y S D I R 0 SYSDIR

6 Y S D I R o 4

gge 1 SDI 1ged 0 others 4

Page 7: Jordan University of Science and Technology Exscind: A ...munzer/courses/INCS745/Lec/Exscind.pdfJordan University of Science and Technology 11 Monther Aldwairi and Duaa Alansari Exscind:

Jordan University of Science and TechnologyJordan University of Science and Technology

7

Related Work

� Mike Fisk et al. proposed a new hybrid system for pattern matching for IDS. This system provide a combination of Boyer-Moore algorithm and Aho-Corasick algorithm as a hybrid system

� Kostas Anagnostakis et al. proposed a new exclusion filter for IDS called E2xB. This stage marks the corresponding cell on the (256-cell) map for each character appearing in the text. Then, the decision that the text contains a character will be only if the corresponding cell of the map is marked with the character will be only if the corresponding cell of the map is marked with the num of the packet

� Sarang Dharmapurikar et al. proposed a new hardware-based technique forBloom filters. They divided the signatures into groups according to their lengthand stored each group in a unique Bloom filter. Each Bloom filter used ananalyzer for probing detected strings to determine whether it is a false positiveor not

� Haoyu Song et al. proposed Extended Bloom Filter (EBF) to implement multi-pattern signature matching. They introduced a new technique to support longsignature matching to minimize the maintained signature lengths. Theyembedded multi-port memories in FPGAs

Exscind @ NWeSP ‘11 by Monther Aldwairi and Duaa Alansari 10/14/2011

Page 8: Jordan University of Science and Technology Exscind: A ...munzer/courses/INCS745/Lec/Exscind.pdfJordan University of Science and Technology 11 Monther Aldwairi and Duaa Alansari Exscind:

Jordan University of Science and TechnologyJordan University of Science and Technology

88

Outline8

� Introduction to Intrusion Detection

� The Problem

� Related Work

� Exscind: The Big Picture� Exscind: The Big Picture

� Bloom Filters

� Exscind Example

� Experimental Evaluation

� Conclusions

Exscind @ NWeSP ‘11 by Monther Aldwairi and Duaa Alansari 10/14/2011

Page 9: Jordan University of Science and Technology Exscind: A ...munzer/courses/INCS745/Lec/Exscind.pdfJordan University of Science and Technology 11 Monther Aldwairi and Duaa Alansari Exscind:

Jordan University of Science and TechnologyJordan University of Science and Technology

9

Exscind the Big picture

� We propose Exscind: meaning to exclude from the union

o Exscind is an exclusion-inclusion intrusion detection

o It can exclude as many clean packets as possible

o Modified Exclusion-Inclusion Bloom filter using only 4-gram

� Produces a probable matches list� Produces a probable matches list

� Provides an index for the first match position

o Modifies the WM

� Search ONLY the probable matches list

� Start from the first index of probable match

� Support all ASCII codes

� Extended, special characters and the Null character

Exscind @ NWeSP ‘11 by Monther Aldwairi and Duaa Alansari 10/14/2011

Page 10: Jordan University of Science and Technology Exscind: A ...munzer/courses/INCS745/Lec/Exscind.pdfJordan University of Science and Technology 11 Monther Aldwairi and Duaa Alansari Exscind:

Jordan University of Science and TechnologyJordan University of Science and Technology

10

Exscind the Big picture

Incoming packets

Exclusion-Inclusion Filter

(Modified Bloom Filter)(Modified Bloom Filter)

Modified Wu-Manber (MWM)

1. Search for the probable matches

2. Search from the first probable match index

Exscind @ NWeSP ‘11 by Monther Aldwairi and Duaa Alansari 10/14/2011

Probable Matching

First Match Index

Page 11: Jordan University of Science and Technology Exscind: A ...munzer/courses/INCS745/Lec/Exscind.pdfJordan University of Science and Technology 11 Monther Aldwairi and Duaa Alansari Exscind:

Jordan University of Science and Technology 11Start

Initializing the modified WM

Program first n-gram of signatures into Bloom vector & storing the SID

For each packet, sliding n-gram window

Check if any

2-hash bits are set ?

No

Yes

Store signature in highly probable list

Empty List ?

Use modified WM to search

Match found ?

Printing match found

End

Yes

No

No

Yes

Page 12: Jordan University of Science and Technology Exscind: A ...munzer/courses/INCS745/Lec/Exscind.pdfJordan University of Science and Technology 11 Monther Aldwairi and Duaa Alansari Exscind:

Jordan University of Science and TechnologyJordan University of Science and Technology

Bloom Filters12

� Exclusion filter through approximate membership queries on a set ofstrings

o Bloom Vector: array vector of m bits

o Programming: use k hash functions on signatures and set the corresponding bits

Querying: Compute the same k hash functions on the packet, then o Querying: Compute the same k hash functions on the packet, then check corresponding hash bits?

� If at least one bit is not set, 100% certainty the signature does not belong

� If all k bits are set, the signature might belong to the set with a probability

o Assume the signature “cmd.exe” hashes to 2 and 8

o Set bits 2 and 8 in a 10-bit Bloom filter

Exscind @ NWeSP ‘11 by Monther Aldwairi and Duaa Alansari 10/14/2011

0 0 1 0 0 0 0 0 1 0

cmd.exe

Page 13: Jordan University of Science and Technology Exscind: A ...munzer/courses/INCS745/Lec/Exscind.pdfJordan University of Science and Technology 11 Monther Aldwairi and Duaa Alansari Exscind:

Jordan University of Science and TechnologyJordan University of Science and Technology

13

Exscind Example

1. Initialize modified WMwith all Snort signatures

2. Program the Bloom vector with the 4-gram prefix of all Snortsignatures

3. Store the SID of each 4-gram corresponding to set bits

Exscind @ NWeSP ‘11 by Monther Aldwairi and Duaa Alansari 10/14/2011

Page 14: Jordan University of Science and Technology Exscind: A ...munzer/courses/INCS745/Lec/Exscind.pdfJordan University of Science and Technology 11 Monther Aldwairi and Duaa Alansari Exscind:

Jordan University of Science and TechnologyJordan University of Science and Technology

14

Exscind Example

4. Slide a 4-gram window throughout the packet

5. Computes the same two hash functions

Exscind @ NWeSP ‘11 by Monther Aldwairi and Duaa Alansari 10/14/2011

Page 15: Jordan University of Science and Technology Exscind: A ...munzer/courses/INCS745/Lec/Exscind.pdfJordan University of Science and Technology 11 Monther Aldwairi and Duaa Alansari Exscind:

Jordan University of Science and TechnologyJordan University of Science and Technology

15

Exscind Example

4. Check the hash bits of each window against the signatures Bloomvector.

5. If they are set, this means that there is a probable match

6. The system retrieves the stored SIDs and adds the signatures to theprobable signatures list.probable signatures list.

Exscind @ NWeSP ‘11 by Monther Aldwairi and Duaa Alansari 10/14/2011

Page 16: Jordan University of Science and Technology Exscind: A ...munzer/courses/INCS745/Lec/Exscind.pdfJordan University of Science and Technology 11 Monther Aldwairi and Duaa Alansari Exscind:

Jordan University of Science and TechnologyJordan University of Science and Technology

16

Exscind Example

8. Exscind checks the probable signatures list

a. If empty � packet is clean � 100% certainty it has no attacksignatures�skip pattern matching

b. If not empty � packet is suspicious � modified WM �

starting from first position of probable match �search only theprobable signatures

c. If any match occurs within the packet, the system considersmalicious packet and records the number of matchesoccurrences along with the index of each occurrence

Exscind @ NWeSP ‘11 by Monther Aldwairi and Duaa Alansari 10/14/2011

Page 17: Jordan University of Science and Technology Exscind: A ...munzer/courses/INCS745/Lec/Exscind.pdfJordan University of Science and Technology 11 Monther Aldwairi and Duaa Alansari Exscind:

Jordan University of Science and TechnologyJordan University of Science and Technology

17

Exscind Savings and Overhead

� Exscind reduces execution timeo Exclusion-Inclusion skips pattern matching for clean packets

o MWM searches suspicious packets ONLY for small probable signatures matching list

MWM search starts from the index for the first probable o MWM search starts from the index for the first probable match

� Exscind over head on memory and timeo Time for BF programming added to preprocessing

o Time for BF querying

o Memory space for Bloom vector and the of signatures SIDs

Exscind @ NWeSP ‘11 by Monther Aldwairi and Duaa Alansari 10/14/2011

Page 18: Jordan University of Science and Technology Exscind: A ...munzer/courses/INCS745/Lec/Exscind.pdfJordan University of Science and Technology 11 Monther Aldwairi and Duaa Alansari Exscind:

Jordan University of Science and TechnologyJordan University of Science and Technology

1818

Outline18

� Introduction to Intrusion Detection

� The Problem

� Related Work

� Exscind: The Big Picture� Exscind: The Big Picture

� Bloom Filters

� Exscind Example

� Experimental Evaluation

� Conclusions

Exscind @ NWeSP ‘11 by Monther Aldwairi and Duaa Alansari 10/14/2011

Page 19: Jordan University of Science and Technology Exscind: A ...munzer/courses/INCS745/Lec/Exscind.pdfJordan University of Science and Technology 11 Monther Aldwairi and Duaa Alansari Exscind:

Jordan University of Science and TechnologyJordan University of Science and Technology

19

Experimental Setup and Metrics

� Metrics1. Average execution time

2. Average memory usage

3. WMSC (Wu-Manber Skip Counter) means the number of 3. WMSC (Wu-Manber Skip Counter) means the number of clean packets skipped without pattern matching

� Experimental Environmento Workstation: 3 GHz Intel Core ™ 2, 4 GB of RAM

o execution time � ExecutionStopWatch class

o memory usage � Garbage Collection class

o 20 experiments average numbers reported

Exscind @ NWeSP ‘11 by Monther Aldwairi and Duaa Alansari 10/14/2011

Page 20: Jordan University of Science and Technology Exscind: A ...munzer/courses/INCS745/Lec/Exscind.pdfJordan University of Science and Technology 11 Monther Aldwairi and Duaa Alansari Exscind:

Jordan University of Science and TechnologyJordan University of Science and Technology

20

Export Packet Contents

� Wireshark Network Protocol Analyzer v1.4.4

Exscind @ NWeSP ‘11 by Monther Aldwairi and Duaa Alansari 10/14/2011

Page 21: Jordan University of Science and Technology Exscind: A ...munzer/courses/INCS745/Lec/Exscind.pdfJordan University of Science and Technology 11 Monther Aldwairi and Duaa Alansari Exscind:

Jordan University of Science and TechnologyJordan University of Science and Technology

21

Packets Extraction

� C# utility to parse payload

Exscind @ NWeSP ‘11 by Monther Aldwairi and Duaa Alansari 10/14/2011

Page 22: Jordan University of Science and Technology Exscind: A ...munzer/courses/INCS745/Lec/Exscind.pdfJordan University of Science and Technology 11 Monther Aldwairi and Duaa Alansari Exscind:

Jordan University of Science and TechnologyJordan University of Science and Technology

22

Signatures Extraction

� Snort 2.9.0.4 rules released in March 2011

� 56 rules files � C# to extract SIDs, content and uricontent� concatenate same rule signatures with zero space

Exscind @ NWeSP ‘11 by Monther Aldwairi and Duaa Alansari 10/14/2011

Page 23: Jordan University of Science and Technology Exscind: A ...munzer/courses/INCS745/Lec/Exscind.pdfJordan University of Science and Technology 11 Monther Aldwairi and Duaa Alansari Exscind:

Jordan University of Science and TechnologyJordan University of Science and Technology

23

Signatures Analysis

� 9,945 signatures

700

800

900

1000

1100

Signatures Count

Exscind @ NWeSP ‘11 by Monther Aldwairi and Duaa Alansari 10/14/2011

0

100

200

300

400

500

600

700

1 5 9 13 17 21 25 29 33 37 41 45 49 53 57 61 65 69 73 77 81+

Signatures Count

Signatures Length

Page 24: Jordan University of Science and Technology Exscind: A ...munzer/courses/INCS745/Lec/Exscind.pdfJordan University of Science and Technology 11 Monther Aldwairi and Duaa Alansari Exscind:

Jordan University of Science and TechnologyJordan University of Science and Technology

24

Malicious Traffic Analysis

Ugly

Trace

Total

Packets

Malicious

Packets

Malicious

Packets (%)

WM

Time

Exscind

Time

Speed

up

8 671116 258556 38.5 25.74 15.09 1.71

57 209188 86129 41.2 15.29 9.53 1.60

51 299713 129763 43.3 15.02 9.50 1.58

58 268000 119627 44.6 11.37 7.42 1.53

Bad

Trace

Total

Packets

Malicious

Packets

Malicious

Packets (%)

WM

Time

Exscind

Time

Speed

up

Exscind @ NWeSP ‘11 by Monther Aldwairi and Duaa Alansari 10/14/2011

Trace Packets Packets Packets (%) Time Time up

1 688158 109311 15.9 75.44 23.00 3.28

22 659365 110363 16.7 76.22 22.90 3.33

0 771382 139866 18.1 66.18 20.87 3.17

2 642091 130901 20.4 55.04 19.11 2.88

Good

Trace

Total

Packets

Malicious

Packets

Malicious

Packets (%)

WM

Time

Exscind

Time

Speed

up

VA 3606 54 1 105 17 6.18

GD 9354 175 2 345 63 5.48

LC 29474 1031 3 698 130 5.37

HM 2817 126 4 90 18 5.00

Page 25: Jordan University of Science and Technology Exscind: A ...munzer/courses/INCS745/Lec/Exscind.pdfJordan University of Science and Technology 11 Monther Aldwairi and Duaa Alansari Exscind:

Jordan University of Science and TechnologyJordan University of Science and Technology

25

Speedup

Exscind @ NWeSP ‘11 by Monther Aldwairi and Duaa Alansari 10/14/2011

Page 26: Jordan University of Science and Technology Exscind: A ...munzer/courses/INCS745/Lec/Exscind.pdfJordan University of Science and Technology 11 Monther Aldwairi and Duaa Alansari Exscind:

Jordan University of Science and TechnologyJordan University of Science and Technology

26

Memory in KBs

WM

Memory

Our

Memory

Added

Overhead

Percentage

overhead (%)

Exscind © J.U.S.T. 2011. Monther Aldwairi and Duaa Alansari

Memory Memory Overhead overhead (%)

537,178 537,766.1 588.1 1.095

Page 27: Jordan University of Science and Technology Exscind: A ...munzer/courses/INCS745/Lec/Exscind.pdfJordan University of Science and Technology 11 Monther Aldwairi and Duaa Alansari Exscind:

Jordan University of Science and TechnologyJordan University of Science and Technology

27

Speedup Scaling

Exscind @ NWeSP ‘11 by Monther Aldwairi and Duaa Alansari 10/14/2011

Page 28: Jordan University of Science and Technology Exscind: A ...munzer/courses/INCS745/Lec/Exscind.pdfJordan University of Science and Technology 11 Monther Aldwairi and Duaa Alansari Exscind:

Jordan University of Science and TechnologyJordan University of Science and Technology

28

Memory Scaling

Exscind @ NWeSP ‘11 by Monther Aldwairi and Duaa Alansari 10/14/2011

Page 29: Jordan University of Science and Technology Exscind: A ...munzer/courses/INCS745/Lec/Exscind.pdfJordan University of Science and Technology 11 Monther Aldwairi and Duaa Alansari Exscind:

Jordan University of Science and TechnologyJordan University of Science and Technology

29

Conclusions

1. Average speedup of 3.4 times 2. Normal traffic speedup > 6 times2. Normal traffic speedup > 6 times3. Overhead: 0.11% increase in memory usage4. Speedup is almost constant and undiminished

with increasing number of signatures 5. Memory usage increases linearly

Exscind © J.U.S.T. 2011. Monther Aldwairi and Duaa Alansari

Page 30: Jordan University of Science and Technology Exscind: A ...munzer/courses/INCS745/Lec/Exscind.pdfJordan University of Science and Technology 11 Monther Aldwairi and Duaa Alansari Exscind:

Jordan University of Science and TechnologyJordan University of Science and Technology

30

QsQs

Exscind © J.U.S.T. 2011. Monther Aldwairi and Duaa Alansari

Page 31: Jordan University of Science and Technology Exscind: A ...munzer/courses/INCS745/Lec/Exscind.pdfJordan University of Science and Technology 11 Monther Aldwairi and Duaa Alansari Exscind:

Jordan University of Science and TechnologyJordan University of Science and Technology

31

Signatures Collisions

� Collision are the number of signatures with same hash bits!

� Bloom vector length 700,000 bit

Exscind @ NWeSP ‘11 by Monther Aldwairi and Duaa Alansari 10/14/2011

Page 32: Jordan University of Science and Technology Exscind: A ...munzer/courses/INCS745/Lec/Exscind.pdfJordan University of Science and Technology 11 Monther Aldwairi and Duaa Alansari Exscind:

Jordan University of Science and TechnologyJordan University of Science and Technology

32

WMSC Worst Case Performance

7880828486889092949698

WMSC Percentage Trace 8

Trace 57

Trace 51

5254565860626466687072747678

4 5 6 7 8 9 10

WMSC Percentage

Prefix Length (n-gram)

Trace 58

Trace 1

Trace 22

Trace 0

Trace 2

Exscind @ NWeSP ‘11 by Monther Aldwairi and Duaa Alansari 10/14/2011

Page 33: Jordan University of Science and Technology Exscind: A ...munzer/courses/INCS745/Lec/Exscind.pdfJordan University of Science and Technology 11 Monther Aldwairi and Duaa Alansari Exscind:

Jordan University of Science and TechnologyJordan University of Science and Technology

33

WMSC Best Case Performance

96

97

98

99

100

WMSC Percentage

SIPETEVA

SVATESP

GD

90

91

92

93

94

95

4 5 6 7 8 9 10

WMSC Percentage

Prefix Length (n-gram)

GD

LC

WYM

WHM

Exscind © J.U.S.T. 2011. Monther Aldwairi and Duaa Alansari

Page 34: Jordan University of Science and Technology Exscind: A ...munzer/courses/INCS745/Lec/Exscind.pdfJordan University of Science and Technology 11 Monther Aldwairi and Duaa Alansari Exscind:

Jordan University of Science and TechnologyJordan University of Science and Technology

34

Worst Case Execution Time Comparison

� For n= 6

54.605 53.241 52.692

48.221

40

45

50

55

60

65

Excution Tim

e (s)

WM Time Exscind Time

20.158

12.879 12.655

8.8099.519

6.332 6.364.604

11.002 10.901 11.125 11.812

0

5

10

15

20

25

30

35

40

Trace 8 Trace 57 Trace 51 Trace 58 Trace 1 Trace 22 Trace 0 Trace 2

Excution Tim

e (s)

DEFCON17 Traces

Exscind © J.U.S.T. 2011. Monther Aldwairi and Duaa Alansari

Page 35: Jordan University of Science and Technology Exscind: A ...munzer/courses/INCS745/Lec/Exscind.pdfJordan University of Science and Technology 11 Monther Aldwairi and Duaa Alansari Exscind:

Jordan University of Science and TechnologyJordan University of Science and Technology

Best Case Execution Time Comparison35

630

450

500

550

600

650

700

Excution Tim

es (m

s)

WM Time Exscind Time

� For n= 6

58

89

304

72

39

6 9

38

85

10 50

50

100

150

200

250

300

350

400

SVATESP SIPETEVA GD LC WHM WYM

Excution Tim

es (m

s)

Normal Traffic Traces

Exscind © J.U.S.T. 2011. Monther Aldwairi and Duaa Alansari

Page 36: Jordan University of Science and Technology Exscind: A ...munzer/courses/INCS745/Lec/Exscind.pdfJordan University of Science and Technology 11 Monther Aldwairi and Duaa Alansari Exscind:

Jordan University of Science and TechnologyJordan University of Science and Technology

Worst Case False Alarms Probability36

Additional pattern matching performed due to the fact the BF only inspects n-grams

7

8

9

10

Alarm

Probability (E-5)

Trace 8

Trace 57

0

1

2

3

4

5

6

7

4 5 6 7 8 9 10

False Alarm

Probability

Prefix Length (n-gram)

Trace 51

Trace 58

Trace 1

Trace 22

Trace 0

Trace 2

Exscind © J.U.S.T. 2011. Monther Aldwairi and Duaa Alansari

Page 37: Jordan University of Science and Technology Exscind: A ...munzer/courses/INCS745/Lec/Exscind.pdfJordan University of Science and Technology 11 Monther Aldwairi and Duaa Alansari Exscind:

Jordan University of Science and TechnologyJordan University of Science and Technology

37

Hash Functions

1. Rolling hash: it is used for n-gram sliding windows movethroughout a text. There are two popular rollingfunctions: Cyclic Polynomial and Rabin-Karp. We useCyclic Polynomial because it is faster than Rabin-KarpCyclic Polynomial because it is faster than Rabin-Karp

2. SAX hash: it is also called Shift-add-XOR, because itcombines two functions; addition and XOR with a shiftoperation. It works as follows for each character in thestring:

SAX hash = (hash << 5) + (hash >> 2) + character

Exscind @ NWeSP ‘11 by Monther Aldwairi and Duaa Alansari 10/14/2011