Improvement of PBFT Algorithm Based on CART

Round 1

Reviewer 1 Report

In current state, the paper is very hard to read and follow the given explanations. The content layout is one of the reasons, as well as the poor writing (many sentences are not easy to understand, and require multiple readings and deduction using the other sentences around the problematic ones). Some examples:

- The first sentence in the abstract tries to tell to many things. The part "combined with the characteristics of the alliance chain" seems unconnected to the sentence and brings to confusion. The third sentence in the abstract is given as an algorithm guide style which is not in line with the other sentences in the abstract. The abstract should be rewritten as in current state it looks more like algorithm description.

- The references should be written in proper format, for example, [2]. Currently, the introduction is very hard to read due to wrong format used for references.

- For many abbreviations, the authors have not provided full names (CART, C-PBFT,...)

- Literature overview is too short and not very informative. Few algorithms are listed and described briefly, but the given descriptions do not relate to the proposed algorithm in sense, in which aspects the proposed (by the other authors) improvements of PBFT are not optimal and how the algorithm proposed in this paper overcomes their shortcomings.

- Figure 1 is not properly explained. The improvements of some steps are given, but the proper explanation should start with the explanation of the flow shown in Figure 1, and then explanations of improvements inside the steps should be given.

- In section 2.1 there is "?? denotes the importance of the attribute, and 0<??<1. " but this variable is given in (3), while the sentence that contains the given text is given for (2)

- n is used ambiguously throughout the paper

- In algorithm 1 description, there is "Calculate the value of ?? and detect the ???? status;" but these variables are explained in latter section

- In section 2.2, different types of nodes are listed, but there is no definitions for each type.

- Figure 3 is confusing and not explained properly in the text.

- Figure 6 suggests that PBFT is better than C-PBFT

These are just some of the parts of the paper that should be addressed. However, in my opinion it would be better if the authors completely rewrite the paper, taking care of the content layout in terms of explanation flow as well as of the proper editing of English language to avoid confusing sentences and text, and then submit it as a new paper, rather then doing the major revision.

Author Response

Thank you for your help in reviewing my manuscript and giving me your valuable comments. The following is my response.

Point 1: The first sentence in the abstract tries to tell to many things. The part "combined with the characteristics of the alliance chain" seems unconnected to the sentence and brings to confusion. The third sentence in the abstract is given as an algorithm guide style which is not in line with the other sentences in the abstract. The abstract should be rewritten as in current state it looks more like algorithm description.

Response 1: Following your advice, I have rewritten the writing abstract to change the problematic descriptions as indicated by the abstract.

Point 2: The references should be written in proper format, for example, [2]. Currently, the introduction is very hard to read due to wrong format used for references.

Response 2: The first draft had [] citations in the literature, probably due to a formatting problem, which have now all been revised, as shown in Chapter 1.

Point 3: For many abbreviations, the authors have not provided full names (CART, C-PBFT,...)

Response 3: The full name of the algorithm in this paper is the CART algorithm, and C-PBFT is now reflected in the abstract, as shown by the abstract.

Point 4: Literature overview is too short and not very informative. Few algorithms are listed and described briefly, but the given descriptions do not relate to the proposed algorithm in sense, in which aspects the proposed (by the other authors) improvements of PBFT are not optimal and how the algorithm proposed in this paper overcomes their shortcomings.

Response 4: The overview of the literature has been reworked to summarize the literature and describe the strengths and weaknesses. How the algorithm proposed in this paper overcomes the disadvantages is shown in Chapter 1.

Point5: Figure 1 is not properly explained. The improvements of some steps are given, but the proper explanation should start with the explanation of the flow shown in Figure 1, and then explanations of improvements inside the steps should be given.

Response 5: An elaboration of the process has now been added before Figure 1, shown by the second paragraph of Chapter 2.

Point 6: In section 2.1 there is "?? denotes the importance of the attribute, and 0<??<1. " but this variable is given in (3), while the sentence that contains the given text is given for (2)

Response 6: This has now been modified and is shown by equation 3 in section 2.1.

Point 7: n is used ambiguously throughout the paper

Response 7: The symbols for the variable n have now been reviewed throughout the text, and duplicates have been replaced with other letters.

Point 8: In algorithm 1 description, there is "Calculate the value of ?? and detect the ???? status;" but these variables are explained in latter section

Response8：The error has now been corrected and is shown by section 2.1, Algorithm 1.

Point 9: In section 2.2, different types of nodes are listed, but there is no definitions for each type.

Response 9: The definitions of the three different types of nodes have now been added, as shown by 2.2.

Point 10: Figure 3 is confusing and not explained properly in the text.

Response 10: Figure 3 has now been restated and is shown by 2.4.

Point 11: Figure 6 suggests that PBFT is better than C-PBFT

Response 11 : Due to a huge error on my part, I got the two discount labels backwards, which has now been corrected and is shown in Figure 6 from section 3.3.

Reviewer 2 Report

This paper proposes a CART-based PBFT algorithm optimization scheme to solve the problems of a random selection of master nodes and low consensus efficiency of the commonly used PBFT algorithm in the consortium blockchain. In general, I can see the novelty of bringing CART with PBFT. Find the detailed comments below:

Writing:

-          There are some typos and grammatical mistakes.

-          The in-text citations should be within square brackets.

-          Some terminologies have not been defined, such as PoS, PBFT, and PoW,

-          usually, the research gap, contributions and organisation are mentioned at the end of the introduction. Hence, the research gap and contributions have not been identified.

-          -what does C-PBFT refer to? Is it concurrent or CART? Also, CART has not been defined.

Literature review: The literature review is very poor, and the authors have not reviewed if other existing studies considered CART with PBFT. Moreover, is there any other optimisation-based PBFT mechanism?

Methodology and Results:

-          Bringing optimisation with the blockchain system makes computation resources very complex.

-          How does the introduction of weighted impurity variables improve the performance of the Classification and Regression Trees (CART) algorithm in the proposed PBFT optimization algorithm?

-          Can you describe the node scoring mechanism and how it ensures the reliability of consensus nodes?

-          Is there any limitation to this integration?

Author Response

Thank you for your help in reviewing my manuscript and giving me your valuable comments. The following is my response.

Point 1:   There are some typos and grammatical mistakes.

Response 1: Grammatical corrections have now been made throughout the text to correct spelling errors.

Point 2:     The in-text citations should be within square brackets.

Response 2: The first draft had [] citations in the literature, probably due to a formatting problem, which have now all been revised, as shown in Chapter 1.

Point 3:    Some terminologies have not been defined, such as PoS, PBFT, and PoW.

Response 3: PoS, PBFT, and PoW have now been added to the introduction to the literature, shown in Chapter 1.

Point 4:    usually, the research gap, contributions and organisation are mentioned at the end of the introduction. Hence, the research gap and contributions have not been identified.

Response 4: The overview of the literature has been reworked to summarise the literature and describe the strengths and weaknesses. How the algorithm proposed in this paper overcomes the disadvantages is shown in Chapter 1.

Point 5:   what does C-PBFT refer to? Is it concurrent or CART? Also, CART has not been defined.

Response 5: The full name of the algorithm in this paper is the CART algorithm, and C-PBFT is now reflected in the abstract, as shown by the abstract.

Point 6:   Literature review: The literature review is very poor, and the authors have not reviewed if other existing studies considered CART with PBFT. Moreover, is there any other optimisation-based PBFT mechanism?

Response 6: The literature review has been reworked to describe the strengths and weaknesses of scholars' improvements to PBF and to show how they overcame the weaknesses in their own time, as shown in Chapter 1.

Point 7:    Bringing optimisation with the blockchain system makes computation resources very complex.

Response 7: The algorithm operates more efficiently by reducing message passing in the blockchain network through optimisation methods, as shown in Chapter 2.

Point 8:    How does the introduction of weighted impurity variables improve the performance of the Classification and Regression Trees (CART) algorithm in the proposed PBFT optimization algorithm?

Response 8: Weighted impurity variables are used to continuously bifurcate the nodes by dichotomizing them. When nodes are grouped in points, the splitting attribute values are used as scoring criteria.

Point9:   Can you describe the node scoring mechanism and how it ensures the reliability of consensus nodes?

Response 9: Nodes are grouped through a scoring mechanism, and nodes with high trust become consensus nodes to complete the consensus process of the system. The reliability of the consensus nodes is ensured through a penalty mechanism. As shown in section 2.2.

Point10:   Is there any limitation to this integration?

Response 10: The PBFT algorithm's limitation on the number of consensus nodes is not addressed by the chemistry approach, and we will focus our research on addressing this issue in future work. As shown in Chapter 4.

Reviewer 3 Report

Although the proposed algorithm looks interesting, the manuscript has a number of serious drawbacks.

1. Abbreviation CART is not clarified

2. Incorrect references [ ] - ???

3. Citation “10 Na, G.; Chuangming, Z.; C. Yang.; S. Lina., and W. He. “Improvement of PBFT algorithm based on network self-clustering,” Computer Application Research, vol. 38, no. 11, pp. 1– 8, 2021.” is not available in the internet

4. The results do not show significant advantages of the proposed method. Everything is at the level of statistical error.

5. The experiments are described in insufficient detail. The source code of the proposed method is not presented.

6. Not sure if electronics journal is the best fit to this topic.

Author Response

Thank you for your help in reviewing my manuscript and giving me your valuable comments. The following is my response.

Point 1:   Abbreviation CART is not clarified

Response 1: The full name of the algorithm in this paper is the CART algorithm, and C-PBFT is now reflected in the abstract, as shown by the abstract.

Point 2:   Incorrect references [ ] - ???

Response 2: The first draft had [] citations in the literature, probably due to a formatting problem, which have now all been revised, as shown in Chapter 1.

Point 3:  Citation “10 Na, G.; Chuangming, Z.; C. Yang.; S. Lina., and W. He. “Improvement of PBFT algorithm based on network self-clustering,” Computer Application Research, vol. 38, no. 11, pp. 1– 8, 2021.” is not available in the internet

Response 3: The DOI of the cited literature is now attached, as shown in ref 10.

Na, G.; Chuangming, Z.; C. Yang.; S. Lina., and W. He. “Improvement of PBFT algorithm based on network self-clustering,” Computer Application Research, 2021,38(11):3236-3242. DOI : 10.19734/j.issn.1001-3695.2021.03.0098.

Point 4:   The results do not show significant advantages of the proposed method. Everything is at the level of statistical error.

Response 4: The summary of Chapter 3 has now been revised and is shown by Chapter 3, Chapter 4.

Point 5:   The experiments are described in insufficient detail. The source code of the proposed method is not presented.

Response 5: The experiments in Chapter 3 are now reworked as shown in Chapter 3.

Point 6:   Not sure if electronics journal is the best fit to this topic.

Response 6: This article belongs to Blockchain Consensus Algorithms, a blockchain topic I found in the journal.

Reviewer 4 Report

The manuscript by Liu et al entitled "Improvement of PBFT Algorithm Based on CART" proposes the improved PBFT algorithm based on CART. 

The advantage if the study is the brief explanation of the basic idea of the study, the authors are familiar with the problem. However, the main principle, the proposed algorithm itself and its distinction from the original one and the terminology are described insufficiently. Some inconsistencies in the text may be found as well as the main body of the work has to be structurized.  The main comments are the following:

1. The comparison of the original and proposed algorithm has to be performed. 

2. The nodes in their multitude should be described all in one place in the very beginning of this study with the detailed explanation of their functions.  

3. The conclusions are merged with an outlook and even in this form looks too brief. The major benefits of this study remained out of the conclusions.

4. I would suggest you to revisit the Paragraph 2.1. The major attention was taken for the possible typo in the text, which is written not enough clearly and takes a time for reading. 

5. "to divide the samples into multiple samples"

6. "When the node attributes are of the same class, or there is no sample to continue the division, then this node is the root of leaf node" I suppose this is not correct.

7. "The proportion of D1 in is" Is this a typo and A was missed?

8. "∆????(??) denotes the sum of the effects of other attributes on the impurity variable of the attribute" This sentence is better to re-formulate, it may be not correct or the reviewer just does not like the phrases like "the sum of the effects" in the context of describing the index using the "sum" not in a algebraic context.

9. What is f should be defined earlier.

10. "Points greater than or equal to 9..."   This sentence is not clear in the text context. The introductory part of this study may be extended to be more clear for the non-specialists in the blockchain networks.  

11. "Nodes with three different states" Even careful reading of the paragraph did not allow to understand what states you have mentioned.

12. "There is no upper bound on the accumulation of candidate and alternate node points" Is this sentence correct?

13. Authors write "...and the candidate node with the highest points and the node is converted to a consensus node" Please, verify the consistency of the sentence.

14. "Credit rating", "minority-majority idea" Please, verify all the terminology.

15. Please, verify the eq. 5.  It seems that it may be simplified. Besides, I would recommend you to extend the text explaining (n+3) in the denominator.  

16. "The nodes in the network are accurately classified by the CART algorithm  model..." the classes should be given.

17. "The PBFT algorithm requires 2/3 of the voting weights to reach consensus, and this paper adopts the minority-majority idea and sets the voting weight threshold as 1/2 of the total voting weights."  Could you explain how do you explain the origin of the improvements due to the reducing the number of the total voting weights.

Author Response

Thank you for your help in reviewing my manuscript and giving me your valuable comments. The following is my response.

Point 1:  The comparison of the original and proposed algorithm has to be performed. 

Response 1: This paper compares them in terms of latency, throughput and fault tolerance, as shown by Chapter 4.

Point 2:  The nodes in their multitude should be described all in one place in the very beginning of this study with the detailed explanation of their functions.  

Response 2: In this paper, because of the need to identify consensus nodes through a point mechanism in many nodes. Then from the consensus nodes, the master node is elected to carry out the consensus of the system. This is shown by 2.4.

Point 3:  The conclusions are merged with an outlook and even in this form looks too brief. The major benefits of this study remained out of the conclusions.

Response 3: The summary and outlook has been reworked and is shown in Chapter 4.

Point 4:  I would suggest you to revisit the Paragraph 2.1. The major attention was taken for the possible typo in the text, which is written not enough clearly and takes a time for reading. 

Response 4: Section 2.1 has now been reworked to correct a number of inappropriate expressions, as well as misspellings, as indicated by Chapter 2.

Point 5:  "to divide the samples into multiple samples"

Response 5: In this paper, the sample of nodes needs to be continuously divided by dichotomization, and the nodes are divided by the best splitting attribute before the integral grouping can be performed. This is shown in Chapter 2.

Point 6:  "When the node attributes are of the same class, or there is no sample to continue the division, then this node is the root of leaf node" I suppose this is not correct.

Response 6: The proposed description has now been modified and the language reorganized for presentation. It is shown in 2.1.

Point 7:  "The proportion of D1 in is" Is this a typo and A was missed?

Response 7: Due to an oversight on my part the statement was caused by an error, which has now been corrected as shown in 2.1.

Point 8:   "∆????(??) denotes the sum of the effects of other attributes on the impurity variable of the attribute" This sentence is better to re-formulate, it may be not correct or the reviewer just does not like the phrases like "the sum of the effects" in the context of describing the index using the "sum" not in a algebraic context.

Response 8: This has been modified to " represents the sum of the effects of the other attributes of attribute A on the impurity variables", as shown in 2.1.

Point 9:  What is f should be defined earlier.

Response 9: The reviewers did not quite understand it because it may have been presented in the wrong order, but it has been revised. It is shown in 2.2.

Point 10:  "Points greater than or equal to 9..."   This sentence is not clear in the text context. The introductory part of this study may be extended to be more clear for the non-specialists in the blockchain networks.

Response 10: It has been changed to "Consensus nodes need to have a score greater than 9, candidate nodes are those with a score between 8 and 9, and alternate nodes are those with a score less than or equal to 8". This is shown in 2.2.

Point 11:  "Nodes with three different states" Even careful reading of the paragraph did not allow to understand what states you have mentioned.

Response 11: Nodes can become consensus nodes, candidate nodes and alternate nodes in three states, as shown by 2.2.

Point 12:  "There is no upper bound on the accumulation of candidate and alternate node points" Is this sentence correct?

Response 12: This sentence has now been amended as shown in 2.2.

Point 13:  Authors write "...and the candidate node with the highest points and the node is converted to a consensus node" Please, verify the consistency of the sentence.

Response 13: This has been corrected to "convert the node with the highest number of points among the candidate nodes to a consensus node", as shown in 2.2.

Point 14:   "Credit rating", "minority-majority idea" Please, verify all the terminology.

Response 14: The full terminology has now been reviewed and problematic terms have been revised.

Point 15:  Please, verify the eq. 5.  It seems that it may be simplified. Besides, I would recommend you

Response 15: The voting weight formula has now been reviewed to simplify the idea and is shown by 2.3.

Point 16:  "The nodes in the network are accurately classified by the CART algorithm  model..." the classes should be given.

Response 16: As the sample is continuously divided into samples by dichotomy, the subdivision is reached by the best splitting attribute. It can be used as a reference for subsequent grouping of nodes, for example.

Point 17:  "The PBFT algorithm requires 2/3 of the voting weights to reach consensus, and this paper adopts the minority-majority idea and sets the voting weight threshold as 1/2 of the total voting weights."  Could you explain how do you explain the origin of the improvements due to the reducing the number of the total voting weights.

Response 17: The traditional PBFT algorithm requires that 2/3 of the total weight of confirmation messages be collected in order to complete consensus. The integral grouping mechanism ensures the reliability of the consensus nodes and therefore reduces the voting threshold of the consensus nodes. This reduces the bandwidth consumption by network broadcasts to achieve algorithm performance improvement. Shown by 2.3.

Round 2

Reviewer 1 Report

The modifications that are done mostly address the raised concerns. Few more things in the paper should be addressed:

- The explanation added for the Figure 1 does not contain explanation about the punishment mechanism. Please, add it.

- Equation (3) - denominator is i, but this seems to be error. Please, check it and correct it if necessary.

- Equation (5) - I suppose that k should used instead of n in bracket.

- Explanation regarding Figure 3 - under 4 you list Submission phase but that does not correspond to figure like previous three phases. Please use Commit and Reply phases as phase 4 and 5, or at least in figure above these two phases add Submission so a reader can correlate Submission phase to content of the figure.

Author Response

Thank you for your help in reviewing my manuscript and providing valuable comments. My sincere regards, here is my response.

Point 1: The explanation added for the Figure 1 does not contain explanation about the punishment mechanism. Please, add it.

Response 1: An explanation of the penalty mechanism has been added to provide a reasonable exposition of Figure 1. As shown in Chapter 2, see line 97 for details.

Point 2: Equation (3) - denominator is i, but this seems to be error. Please, check it and correct it if necessary. 

Response 2: The denominator characteristic of equation (3) has now been modified as shown in Section 2.1.

Point 3: Equation (5) - I suppose that k should used instead of n in bracket.

Response 3: I did not check for errors due to my oversight. It has now been modified as shown in equation (5) in Section 2.3.

Point 4: Explanation regarding Figure 3 - under 4 you list Submission phase but that does not correspond to figure like previous three phases. Please use Commit and Reply phases as phase 4 and 5, or at least in figure above these two phases add Submission so a reader can correlate Submission phase to content of the figure. 

Response 4: The Commit and Reply phases are now used as phases 4 and 5, and the Submission phase is changed to the Commit phase, which corresponds to the process in Figure 3. As shown in Section 2.4.

Reviewer 3 Report

All in all, this is a pretty interesting study. Issues were fixed. Recommended for publication.

Author Response

Thank you for your recognition of my article and I send my sincere best wishes.