Artificial Intelligence Control Logic in Next-Generation Programmable Networks

Round 1

Reviewer 1 Report

The works presented in the paper are novel and according to the demands of the new era in networking.

I recommend the following revisions to the authors:

Re-write or split the sentence i.e. Automized management of the networks resources in relation ….i.e. lines #78, 79, and 80.

Include some recent literature on the use of AI in SDN for network management e.g. “Management of Software-Defined Networking Powered by Artificial Intelligence” given on the following link:

 https://www.intechopen.com/online-first/76133

In a table show the previous works analysis i.e. in a table mention the limitations of the previous works, whether they used AI or not, whether SDN was used or not etc.

Please show some other results i.e. for E2E delay.

The explanation for figure 3 is not enough please provide more details. Similarly for figure 4.

Why the ONOS Controller is selected for experiments? The ONOS has already the intents logic and intelligence to calculate the routes etc.

 

Author Response

Dear Sir,

We wish to thank for careful reading of the manuscript and for a set of helpful comments. We hope that after the revision the manuscript will find the approval of the editorial board. The changes to the manuscript are listed below. In the text we use brown text to ease the tracking of the changes. It was also necessary to add some references. The details of these changes we provide in the reply to reviewers’ comments below.

Kind regards,
Stanislaw Kozdrowski

Author Response File: Author Response.pdf

Reviewer 2 Report

1. I hope the authors to outline "FlexNet Project" in more detail.

2. In Ch. 2: The problem shoud be defined more clearly. 
   Objective function, decision variables and constraints of the problem should be defined more clearly. 

3. Is this problem first introduced in this paper? Otherwise, provide more comprehensive literature review of research papers on the problem. 

4.  In Ch. 4: I feel the network in Figure 2 is very small. I wonder if AI based approach is needed for such network. In other words, numerical experiments should be performed by using larger network. 

5. Figure 3~6 : Is there any existing management policy very widely used in network management? In order to demonstrate performance of the proposed methodology, I hope the authors to compare their methodology to existing ones. Moreover, results of the comparative experiments should be discussed.

6. Major contributions of this paper should be defined clearly. 

New optimization problem, relevant to network management, is proposed?

New algorithm or methodolody is applied to existing optimization problem?
  
Currently, it is hard to understand what important contribution is provided by this paper. 

 

Author Response

Dear Sir,

We wish to thank you for your careful reading of the manuscript and for a set of helpful comments. We hope that after the revision the manuscript will find the approval of the editorial board. The changes to the manuscript are listed below. In the text, we use brown text to ease the tracking of the changes. It was also necessary to add some references. The details of these changes we provide in the reply to reviewers’ comments are in the attached file.

Kind regards,
Stanislaw Kozdrowski

Author Response File: Author Response.pdf

Reviewer 3 Report

Dear Authors,

 

I give you my extended remarks and suggestions regarding the article "AI Control Logic in Next Generation Programmable Networks".

 

A few remarks and suggestions:

 

• The article is solid structured from summary to conclusion and the whole content makes one meaningful whole without deviating from the main topic,

 

• The research covered in the article propose the solution for network resources allocation, where Artificial Intelligence algorithm is responsible for controlling intent-based routing in Software Defined Networking,

 

• Authors also confirm in the presented results the validity of applied Artificial Intelligence approach to the problem of improving network performance in next-generation networks and usefulness of Networked Application Emulation System traffic generator for efficient economical and technical deploying in Internet of Things networking systems evaluation,

 

• The introduction gives a concise overview of processed scientific field,

 

• Also, introduction indicates the area that the researchers detected, and in which they have contributed concrete scientific contributions,

 

• Notes:

 

• • Perhaps to consider using the complete term (Artificial Intelligence) instead of the abbreviation (AI) in the title of the article,
  • The article mentions several times (for the first time in line 36) the contribution gained through the FlexNet project (this is not in question), however, please add more information about the project and a reference to the same,
  • Line 70 - the abbreviation "IoT" is not defined,
  • Line 90 - the abbreviation for the term Knowledge-Based Network is "KDN", perhaps more appropriate is "KBN" because "KDN" is an abbreviation for Knowledge-Defined Network (go through the text where term is mentioned),
  • Chapter 2. Problem Description - it is mentioned that various tools and techniques were used in the research (ifstat, Deep-Q-Network, Iroko framework, Open Ai Gym, etc.), think about short explanations why it was used and not some other tools and techniques,
  • Sentences in lines 135-137, explain why such a network configuration through Deep-Q-Network (example ...hidden layers of ten, five, five, and three neurons…),
  • Line 211 - the abbreviation "MQTT" is not defined,
  • Line 232 - the abbreviation "UDP" is not defined,
  • Sentence in lines 232-234, remove sentence from brackets,
  • Chapter 4. Experiments and Results - Arrange illustrations 3, 4, 5 and 6 between the text in Chapter 4 to which they refer,
  • Line 302 - typo "taffic" instead of the word traffic,
  • Line 409 - reference 31, missing where / what / how information about the source.

 

• The article gives interesting results and a good introduction to future research with real traffic Internet od Things data and implementation of results in real scenarios,

 

• The article contributes with concrete scientific contribution,

 

• My assessment is that the article be accept after minor revision for publication in a journal.

 

My concluding opinion is that the article is accepted after minor revision for publication!

 

Sincerely

Author Response

Dear Sir,

We wish to thank you for your careful reading of the manuscript and for a set of helpful comments. We hope that after the revision the manuscript will find the approval of the editorial board. The changes to the manuscript are listed below. In the text, we use brown text to ease the tracking of the changes. It was also necessary to add some references. The details of these changes we provide in the reply to reviewers’ comments are in the attached file.

Kind regards,
Stanislaw Kozdrowski

Author Response File: Author Response.pdf

Reviewer 4 Report

The manuscript presents a Deep-Q-Learning based approach for the management of SDNs.

The solution includes components developed in FlexNet project, and more details are welcome regarding their implementation (e.g., a figure depicting the NAPES architecture).

The paper is well written, but there are some aspects that need to be addressed before publishing:
- Fig.1 is captioned AI architecture but looks more like a conceptual model of the overall system. It would be most welcome to include a formal description of the Deep-Q-Learning algorithm that was implemented. In addition, the inclusion of the IMR block in the Figure is not explained.
- there are some minor language issues that must be addressed, including a consistent reference to the FlexNet project (e.g., line 175 was written as FLEXNET).

Author Response

Dear Sir,

We wish to thank you for your careful reading of the manuscript and for a set of helpful comments. We hope that after the revision the manuscript will find the approval of the editorial board. The changes to the manuscript are listed below. In the text, we use brown text to ease the tracking of the changes. It was also necessary to add some references. The details of these changes we provide in the reply to reviewers’ comments are in the attached file.

Kind regards,
Stanislaw Kozdrowski

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report

The authors have revised their paper to some extent.

However, I feel the manuscript is not suitable for publication, yet. 

1) The "novel ML algorithm" is explained only in a single paragraph, "The artificial neural network used in this research is a Deep-Q-Network (DQN) [29] consisting of four inputs (observations), two outputs, and four hidden layers of ten, five,
five, and three neurons. We used DQNs, because it gave the best results in our internal studies. The learning process was  implemented using Ray framework [30] and its goal
was to maximize the total throughput in the network. The neural network takes a state of a network from a viewpoint of a single intent as an input and as an output it produces the expected q-function value for two cases: a) when nothing happens and b) when the path is switched. "

I think that the issues such as training data and parameter settings should be dealt with more carefully. Overall, the descriptions of "novel ML algorithm" is too short. 

2) I think comparison experiment is an important part for this kind of research paper. If it is difficult to compare the proposed algorithm to other existing complex solution methods, the authors can compare the network generated by the proposed algorithm to randomly generated networks.

Moreover, authors can request editorial office to extend due date for revision, if required. 

 

Author Response

Dear Sir,

We wish to thank you for your careful reading of the manuscript and for a set of helpful comments. We hope that after the revision the manuscript will find the approval of the editorial board. The changes to the manuscript are listed below. In the text, we use blue text to ease the tracking of the changes. The details of these changes we provide in the reply to reviewers’ comments are in the attached file.

Kind regards,

Stanislaw Kozdrowski

Author Response File: Author Response.pdf

Round 3

Reviewer 2 Report

I hope adequate follow-up studies would be performed.