Research Studies on Possible Improvements to the Aho-Corasick String Matching Algorithm

Abstract

Four data structures (AC-basic, AC-expanded, AC-bitVec and AC-compressed) are implemented in this project in order to realize Aho-Corasick (AC) algorithm. Statistics on space and time requirements of these four implementations on scanning various input files are collected and compared.

In AC-expanded, a two-dimensional array is used to store the fully expanded transition rule table, which takes up the most memory space among the four. AC-basic uses linked-lists to save memory space, but the processing speed is lowered by this practice. AC-bitVec transforms the linked-lists to bit vectors to lower the time and memory requirement. AC-compressed is a reasonable improvement over the above three, in which the transition rule table is compressed by perfect hashing and elimination of transition edges.

Comparing the results obtained by running these four versions, the processing speed of AC-compressed reaches 3.9 to 7.8 times that of AC-basic, while it is 34% to 83% of that of AC-expanded. Regarding memory space requirement, AC-compressed needs only 3.8% to 6.3% the space of AC-expanded, and 1.7 to 2.6 times of that of AC-basic. Thus, the proposed AC-compressed implementation can be served as a satisfying trade-off between memory and speed over the basic version and expanded version.