140 likes | 355 Views
A Memory Efficient DFA based on Pattern Segmentation for Deep Packet Inspection. Author: Yeim-Kuan Chang and Jo- Ning Yu Publisher : Presenter: Yuen- Shuo Li Date : 2013/04/24. Background.
E N D
A Memory Efficient DFA based on Pattern Segmentation for Deep Packet Inspection • Author: Yeim-Kuan Chang and Jo-Ning Yu • Publisher: • Presenter: Yuen-Shuo Li • Date: 2013/04/24
Background As the role of NIDS has become more important, we have to develop a new high-throughput algorithm to find out the hidden virus in packet payload because the performance of pattern match algorithm is the bottleneck of NIDS.
Method of improving AC • Cutting pattern into sub-patterns (pattern segmentation) • Parallel Match Top k Levels • Bitmap-based compression
Pattern segmentation Backward Transitionscan avoid repeat matching with the same sub-pattern. It can improve the performance of match process. backward Transitions
Pattern segmentation(cont.) 10 states 16 states
Parallel Match Top k Levels The transitions going back to one of the top k levels account for a very large proportion of all transitions.
Parallel Match Top k Levels To reduce memory usage, we adopt the parallel architecture to remove these transitions.
h e r s Bitmap-based compression 1 2 8 0 9 i s 7 6 r s h e 4 5 3
Input stream : h e x r o s e Pattern set : { heroes, rose, hohero } Pattern set’ : { he, ro,es, se, ho } e sub pattern match FSM Main optimized AC automaton (Optimized AC automata) 2 1 h o 5 0 r o 3 4 s e e 7 6 s 9 8 11