Open AccessParallel multiple pattern matching schemes based on cuckoo filter for deep packet inspection on graphics processing units
Author(s) -
Ho ThienLuan,
Cho SeongJe,
Oh SeungRohk
Publication year - 2018
Publication title -
iet information security
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.308
H-Index - 34
eISSN - 1751-8717
pISSN - 1751-8709
DOI - 10.1049/iet-ifs.2017.0421
Subject(s) - bloom filter , computer science , filter (signal processing) , cuckoo , network packet , cuckoo search , parallel computing , matching (statistics) , throughput , set (abstract data type) , deep packet inspection , algorithm , computer network , computer vision , telecommunications , zoology , statistics , mathematics , particle swarm optimization , wireless , biology , programming language
A large amount of data now being transferred through networks has made deep packet inspection (DPI) an essential part of security activities. Several DPI systems are developed based on Bloom filters to defend against malicious worm attacks through the Internet. These approaches have achieved significant performance. However, they do not permit deletion of items from the set of target patterns. This study proposes two multiple pattern matching schemes for DPI to exploit high parallelism capacity of graphics processing units (GPUs). Firstly, a GPU‐based Cuckoo filter scheme is proposed by adopting a new approximate set membership, called Cuckoo filter, for parallel multiple pattern matching. The Cuckoo filter has many advantages over the Bloom filter such as higher insert performance, higher lookup throughput, less memory consumption, less false positive rate, and delete operation support. Secondly, an implementation of the GPU‐based Cuckoo filter, called GPUshared‐based Cuckoo filter is proposed. This scheme can efficiently distribute input string and pre‐processing data in the hierarchical memory of GPUs to optimise the performance of the GPU‐based Cuckoo filter scheme. Experiments show that the proposed schemes offer better performance than the previous approaches based on the Bloom filter.