Accelerating Pattern Matching Using a Novel Multi-Pattern-Matching Algorithm on GPU
Round 1
Reviewer 1 Report
In the paper, an algorithm is proposed, and compared with the actual commonly used Aho Corasick and Wu Manber ones through experiments, which proves it is more efficient than the above two ones in the DPI system. Experiments is comprehensive, and results are sufficient and can be compared with other popular algorithms to prove the usability.
A novel algorithm with better performance has been proposed in the area that has been studied for a long time.
Formatting issues exist in equations.
The first c of Aho-Corasick in the caption of Figure 4 is not capitalized.
The word 'proposed algorithm' in the article can be referred to as PA when used, which can improve the readability of the paper.
Minor editing of English language required.
Author Response
Please see the attachment.
Author Response File: 
Author Response.docx
Reviewer 2 Report
I think this is an interesting work that primarily discusses a novel Deep Packet Inspection (DPI) pattern matching algorithm optimizing memory usage and processing speed. The manuscript presents a clear structure and provides detailed descriptions of the algorithm's detail, enabling readers to understand your intentions. The experimental comparisons with traditional automaton-based algorithms, such as Aho-Corasick and Wu-Manber, demonstrate the advantages of the proposed algorithm in terms of memory usage and processing time during the DPI pattern matching process. However, there are still some aspects that can be improved:
1. While the manuscript focuses on describing the algorithm's detail, the description of the GPU parallelization process in section 5.1 is limited to a brief explanation of the memory migration process. I recommend providing a more detailed flowchart illustrating the integration of the proposed algorithm with the GPU in the context of the DPI system.
2. In the Introduction, the authors mention the contribution of applying optimization techniques to overcome difficulties arising from pattern-matching algorithms during the execution of the DPI process on the GPU platform. Although the GPU-related limitations are specified in section 4, the specific measures taken to address these issues in the DPI system running on the GPU are not clearly described. It would be beneficial to provide more details on how these challenges were overcome.
3. While the manuscript extensively compares the related algorithms in terms of memory usage and throughput, there is a lack of temporal comparisons. I recommend conducting additional experiments to compare the execution time of the proposed algorithm against the other algorithms to provide a more comprehensive performance evaluation.
4. The formatting of formulas and tables should be improved by center-aligning them. Additionally, the resolution of the figures, especially Figure 3 and Figure 12, needs to be increased to ensure clarity.
It is commendably clear and allows for easy comprehension of the content. You have effectively communicated your ideas and findings, making the manuscript highly accessible to a wide readership. However, some statements can be further simplified. For example, the sentences similar to the 2.3.1, "This section may be divided by subheadings. It should provide a concise and precise description of the experimental results, their interpretation, as well as the experimental conclusions that can be drawn" can be revised for conciseness.
Author Response
When we convert the Word version to the pdf version, the resolution of Figure 15 decreases. Due to time constraints, I haven't been able to deal with this problem yet.
Please see the attachment.
Author Response File: Author Response.docx
Reviewer 3 Report
The authors present a new multi-pattern matching algorithm, which reduces the memory space and time in the DPI pattern-matching. They first provide deep background about the algorithms. Then they propose the problem statement and provide the solution. Finally, they conduct the experiment and compare the pattern-matching between different algorithms. The proposed algorithms show much-improved performance in memory space and time. I think this work is data-rich and well-written. I do not have major concerns about the quality of this work. I have one more suggestion:
How is the accuracy of the proposed algorithm compared with the conventional method? Any improvement?
Author Response
Please see the attachment.
Author Response File: Author Response.docx
Round 2
Reviewer 2 Report
The revised version looks fine. Good work!
