Variable-Stride Multi-Pattern Matching For Scalable Deep Packet Inspection

Variable-Stride Multi-Pattern Matching For Scalable Deep Packet Inspection

Author: Nan Hua, Haoyu Song, T. V. Lakshman

Publisher:INFOCOM 2009

Presenter: Chun-Yi Li

Date: 2009/04/22

2

Related Work Winnowing Algorithm

Variable-Stride DFA

Algorithm Optimizations

Performance

Outline

3

Related Work

s e r v e r [ p a t h ]

s e r v e r [ p a t h ]

177 87 210 119 87 178 63 92 57 56 71

Winnowing with k = 2 and w = 3

Winnowing Algorithm

1. Calculate the hash value of every consecutive k characters.2. Use a sliding window of size w to select the minimum hash value in

the window.A tie is broken by selecting the rightmost minimum value.

delimiter

4


Variable-Stride DFA


Performance

Outline

5

Variable-Stride DFA

Segmentation Scheme Properties

Property 1: • The size of any segmented block is in the range [1, w].• Tail block sizes are in the range [k−1, w+k−2].• Indivisible pattern sizes are in the range [1, w+k − 2].• Coreless pattern sizes are in the range [w+k−1, 2w+k−2].

Coreless patternIndivisible pattern

6

Variable-Stride DFA

Segmentation Scheme PropertiesProperty 2:

If a pattern appears in a data stream then segmenting the data stream results in exactly the same delimiters for the core blocks of the pattern.

The head block can be affected by the preifix and the tail block can be affected by the suffix. However, the core blocks are totally confined to the pattern and isolated from the context.

ex: input stream: ...A|BCh|ij|kl|m|nD|EF|...pattern: hij|kl|mn

7

Variable-Stride DFA

Finite Automaton Construction

patternhead string core string

tail string

s1 ridiculous r id ic ulo u s

s2 authenticate auth ent ica te

s3 identical id ent ica l

s4 confident conf id ent

s5 confidential conf id ent ial

s6 entire ent (empty) ire

s7 set --- (indivisible) ---

quasi-match state

8

Variable-Stride DFA

System Design and Basic Data Structure

9

Variable-Stride DFA


State Transition Table(STT) Match Table(MT)

Hash Key Value

Start State block End

State

q0 id q14

q0 ent q1

q14 ic q2

q2 ulo q3

q3 u q11

q14 ent q15

q1 ica q12

q15 ica q12

Match TableState Head Tail Depth

q11r s 4

q12auth te 2

q12id l 2

q14conf ent 1

q15conf ial 2

10

Variable-Stride DFA


Head Queue(HQ)

w bytes

D entries(D is the length of the longest forwarding path of the VS-DFA)

To enable match verification on the Quasi-match states, we need to maintain a Head Queue (HQ) that remembers the Block-matching history.

11

Variable-Stride DFA


Match Table

State Head Tail Depth

q11 r s 4

q12 auth te 2

q12 id l 2

q14 conf ent 1

q15 conf ial 2

ex:

Data Stream: ‥‥A|BCr|id|ic|ulo|u|sD|EF‥‥

Head Queue(HQ)

u l o u

c u l o

i d i c

C r i d

A B C r

01234

12

Variable-Stride DFA


Match Table

State Head Tail Depth

q11 r s 4

q12 auth te 2

q12 id l 2

q14 conf ent 1

q15 conf ial 2

ex:

Data Stream: ‥ ‥ABCD|Eau|th|ent|ica|te ‥ ‥

Head Queue(HQ)

t i c a

h e n t

a u t h

D E a u

A B C D

01234

13

Variable-Stride DFA

Short Pattern Handling

Headw byte

Tailw+k-2 byte

* e n t i r e *s e t * * * * ** s e t * * * ** * s e t * * ** * * s e t * *

Coreless Pattern

Indivisible Pattern

Using TCAM for short pattern lookups

14


Variable-Stride DFA


Performance

Outline

15


Reducing Single-Byte Blocks

It is possible to generate specific inputs that result in only single-byte streams being produced independent of the chosen hash functions and window parameters.

16


Combination Rule 1 (applied on data stream)

‥‥ c1 c2 c3 c4 c5 c6 c7 c8 c9 ‥‥

‥‥ c1 c2 c3 c4 c5 c6 c7 c8 c9 ‥‥

w = 3

17


Combination Rule 1 (applied on pattern)

c1 c2 c3 c4 c5 c6 c7 c8 c9 c10 c11 c12 c13 c14 c15


window size w = 3

Step 1:

18



1.

2.

3.

4.

5.

6.





window size w = 3




Step 2:

Replicate

19


Combination Rule 1 (applied on pattern)data stream:

pattern 1:


a1 a2 c1 c2 c3 c4 c5 c6 c7 c8 c9 c10 c11 c12 c13 c14 c15

Match

window size w = 3


Applying Combination Rule 1

20



pattern 2:


‥ a1 c1 c2 c3 c4 c5 c6 c7 c8 c9 c10 c11 c12 c13 c14 c15

Match

window size w = 3



21



pattern 3:


‥‥ c1 c2 c3 c4 c5 c6 c7 c8 c9 c10 c11 c12 c13 c14 c15

Match

window size w = 3



22



pattern 4:



Match

window size w = 3



23



pattern 5:



Match

window size w = 3



24



pattern 6:



Match

window size w = 3



25


Combination Rule 2 (applied on data stream)

‥‥ c1 c2 c3 c4 c5 c6 c7 c8 c9 ‥‥

‥‥ c1 c2 c3 c4 c5 c6 c7 c8 c9 ‥‥

window size w’= w+1 = 3+1 = 4


26






Step 1:

27






Replicate


1.

2.

Step 2:

28





Match

‥c1 c2 c3 c4 c5 c6 c7 c8 c9 c10 c11 c12 c13 c14 c15 a1 ‥

‥c1 c2 c3 c4 c5 c6 c7 c8 c9 c10 c11 c12 c13 c14 c15 a1 ‥


pattern 1:

29





Match

‥c1 c2 c3 c4 c5 c6 c7 c8 c9 c10 c11 c12 c13 c14 c15 ‥ ‥

‥c1 c2 c3 c4 c5 c6 c7 c8 c9 c10 c11 c12 c13 c14 c15 ‥ ‥


pattern 2:

Three STTs Design


Hash Key Value

id q’1

ent q’6

q’1 (q14) ->s4

q’2 (q2) uloq’3 (q3) uq’4 (q11) ->s1

q’5 (q15) ->s5

q’6 (q1) icaq’7 (q12) ->s2

Hash Key Value

q’1 ic q’2

q’1 ent q’4

q’5 ica q’6

Start STT Main STT Jump STT

31


Variable-Stride DFA


Performance

Outline

Performance

Mem1 denotes the memory consumed by “Start STT”Mem2 denotes that for “Three STT”.

Performance

Fixed:patterns extracted from the fixed string rules.Full: the expanded pattern sets that also include the fixed strings extracted from the regular expression rules.

Performance

SNORT-fixed ClamAV-fixed

Performance

Variable-Stride Multi-Pattern Matching For Scalable Deep Packet Inspection

Documents