Large-Scale Dynamic Graph Processing with Graphic Processing Unit-Accelerated Priority-Driven Differential Scheduling and Operation Reduction

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

The manuscript titled "Large Dynamic Graph Processing with GPU-Accelerated Priority-Driven Differential Scheduling and Operation Reduction" presents an interesting and well-structured study. The proposed method combines priority-driven scheduling and operation reduction techniques to optimize dynamic graph processing on GPUs. The research demonstrates significant performance improvements, showing up to 348% faster execution compared to Subway and 120% faster than EGraph. The experiments are well-designed, and the conclusions are supported by strong results.

The paper is relevant to the field of GPU computing and graph processing, and the findings can be useful for researchers working on large-scale parallel processing. The introduction provides a clear background, and the methodology is logically structured. The experimental setup is appropriate, and the results effectively demonstrate the benefits of the proposed method.

While the paper is of high quality, some aspects could be improved to make it clearer and more informative. Below are the key remarks for improvement:

1. Line 216-224: The explanation of the priority-driven scheduling function (Equation 2) is not entirely clear. A short pseudocode or a step-by-step explanation should be added to clarify how priorities are assigned and updated dynamically.

2. The manuscript states that EGraph does not optimize for preliminary active vertices, but this claim is not clearly supported by a direct comparison. A small table comparing the key features of Subway, EGraph, and the proposed method would highlight the differences more effectively.

3. The manuscript claims that the proposed method reduces memory transfers between CPU and GPU, but there is no quantitative evidence for this in the results section. A table or graph with memory transfer statistics would strengthen this argument.

The manuscript is a valuable contribution to GPU-based graph processing. The method is well-designed, and the results confirm its efficiency. However, addressing these key points will further improve clarity, support the claims with stronger evidence, and make the comparisons with existing methods more transparent.

With these minor refinements, the paper will be even stronger and ready for publication.

Author Response

Dear Reviewer,

We would like to sincerely thank you for your attentive indications and good comments.

Our paper is partially rewritten and complemented in order to reflect your comments.

Please refer to the attached revision note and the revised paper.

Many thanks.

Jaesoo Yoo

Author Response File: Author Response.docx

Reviewer 2 Report

Comments and Suggestions for Authors

-    Eqn. 1 seems to consider that all the scheduled partitions have the same maximum allowed size which is the right-hand-side of  (1).  Can this maximum to  be adapted  for each individual partition?. That is: Cardinal(Global) = Sum ( Cardinal (Partition P(i) ) *N(i)) ;   N= Sum ( N(i)) ?.
-    Eqn. 2 has  to be explained more clearly and, may be, rewritten in a more clear way. For instance, S and S/subt are not explained. The summation in the numerators of the right-hand-side are on S/sub t but then S/subt does not appear anyway in the summed-up amounts either as variables or as indexing terms.
-    Do the graph vertices to be linked in some way to Eqn. (2)?. They are not visualized in the equation but, however, they are mentioned in the paragraph which follows to the equation and they are also crucial in the displayed Fig. 4.
-     Is there some updating criterion ( leading t  fix K) which could  be  evaluated analytically. That is, it can occur a dynamic  “updating”, and it is quantified in (2) via “K”,  but it is not explained in the updating process  can be decided via some analytic evaluation formula.
-    The graph theory is not invoked in the paper. For instance, it is not clear if loops can exist (i.e. transitions from a vertex to itself though other intermediate vertices or edges closed on individual vertices) .
-    The section of conclusions is ambiguous in the sense that there are not precise details of what has been done in the research.
-    There are some formatting mistakes in the article pdf in the list of Abbreviations previous to the list of references.

Author Response

Dear Reviewer,

We would like to sincerely thank you for your attentive indications and good comments.

Our paper is partially rewritten and complemented in order to reflect your comments.

Please refer to the attached revision note and the revised paper.

Many thanks.

Jaesoo Yoo

Author Response File: Author Response.docx

Reviewer 3 Report

Comments and Suggestions for Authors

The author must realize the following improvements:

1. The conclusions are poor. These must high-lighted the technical contribution and give the advantages on others or another schemes or models of dynamic graphs used in the process of redundant routines and optimization in transmissions cost of a communication system or managing data volume.  The author must include precise contributions with variables and parameters. The author is obligated to give a prospective and future perspective of the dynamic graph model used. 
2. The introduction must include relations with hardware and software precise surveys and diverse developmentsfollowed in the actuality and the pretension of the improvement in the big data managing and web of wide spectrum. 
3. The author must correct some typographic errors that appear in some parts of the paper, for example in the abbreviations. 

Author Response

Dear Reviewer,

We would like to sincerely thank you for your attentive indications and good comments.

Our paper is partially rewritten and complemented in order to reflect your comments.

Please refer to the attached revision note and the revised paper.

Many thanks.

Jaesoo Yoo

Author Response File: Author Response.docx

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

Dear Authors,

The detailed point-by-point response to the review comments is greatly appreciated. The revised manuscript has been carefully examined, and the improvements made are evident. The clarifications provided effectively address the concerns raised, enhancing the overall quality of the work.

After a thorough evaluation, the manuscript is considered well-structured, clear, and scientifically sound. No further revisions are deemed necessary, and it is found to be suitable for publication in its current form.

Reviewer 2 Report

Comments and Suggestions for Authors

There are no  further comments on the revised version,

Reviewer 3 Report

Comments and Suggestions for Authors

none