Round 1

Reviewer 1 Report

In this paper an interesting topic was discussed in order to increase the power consumption savings in mobile edge computing. In general, the presentation of the paper is good, but it becomes cumbersome from section 4.1.

My suggestions/recommendations are below:

First of all, I recommend to check the cross references of Figures and Tables. For instance, in the manuscript there is a Figure 3 (Probability of the emit interval in different configurations) but no information about this one was provided.

I recommend to make a bullet list related the novelty and the scientific contribution of the paper, in order to point out better it. In the present form I think it is dispersive.

In section 4 the Authors introduced and only briefly discussed Figure 2 but the real explanation of this figure is provided in section 4.1 after describing the active and sleep modes of nodes. I think it could be better to join the arguments.

Figure 3: no explanation of this figure is provided. Is it necessary? If no, please delete it.

Table 4: no cross reference can be found. At row 342, is the Table 3 Table 4?

The quality of the figures is poor, please improve it.

Finally, I suggest to create a new section “methodology” where describe the methodology adopted in the paper. I think that more explanation and information about this could be useful for the reader.

 

Best Regards

Author Response

Reviewer: 1

1. Short Summary: 

In this paper an interesting topic was discussed in order to increase the power consumption savings in mobile edge computing. In general, the presentation of the paper is good, but it becomes cumbersome from section 4.1.

Response summary

We would like to thank the reviewer for carefully reviewing our paper. Many constructive comments are offered. We have revised our paper according to your suggestions. We provide point-to-point responses to each of your comments.

2. Weaknesses: 

(1) First of all, I recommend to check the cross references of Figures and Tables. For instance, in the manuscript there is a Figure 3 (Probability of the emit interval in different configurations) but no information about this one was provided.

Response:

 Thank you for offering this recommendation. We highly appreciate your constructive suggestion and revise this problem as follows.

Figure 1 shows the edge computing network composition architecture, so we mention it in section 2.1.  

Figure 2 shows the probability of the emit interval in different configurations. We mention it in subsection 2.1.2 and make a detailed description.

Figure 3 shows the power consumption of the mobile device in different configurations, which we use this figure to represent the difference of the edge devices. So, we mention it in section 4.1.

Figures 4 to 8 are the experimental results, which have been described in our manuscript.

(2) I recommend to make a bullet list related the novelty and the scientific contribution of the paper, in order to point out better it. In the present form I think it is dispersive.

Response:

Thank you very much for raising this issue, and we have listed our contribution in lines 42 to 50.

(3) In section 4 the Authors introduced and only briefly discussed Figure 2 but the real explanation of this figure is provided in section 4.1 after describing the active and sleep modes of nodes. I think it could be better to join the arguments.

Response:

Thank you very much for raising this issue, and we have revised the label and position of this figure.

(4) Figure 3: no explanation of this figure is provided. Is it necessary? If no, please delete it. Table 4: no cross reference can be found. At row 342, is the Table 3 Table 4?

The quality of the figures is poor, please improve it.

Response:

Thank you for offering this recommendation. We change the number and position of figure 3 which is mentioned in response (3).

We revise the reference of Table 3 in lines 248 and renamed it as Table 2.

The table mentioned in line 342 is Table 4, and we have revised it.

(5) Finally, I suggest to create a new section “methodology” where describe the methodology adopted in the paper. I think that more explanation and information about this could be useful for the reader.

Response:

Thank you for offering this recommendation. Due to the limitation of paper length and revision time, we haven’t added this part in our manuscript. We will complete the “methodology” in our future work.

 

Author Response File: Author Response.docx

Reviewer 2 Report

explain all abbreviations the first time they are used

spellcheck including punctuations and blanks

table 1 popps out of the blue, why is it where it is and what does it tell us?

the algorithms need to be presented in tables or figures and referred to, instead of pseudo code  I would appreciate a flowchart

where does table 2 belong to?

how is data management considered in the model presented?

to assess the relevance of the simulation and findings: can you give examples for the different tasks and relate to savings

 

The conclusions are brief, contain several language errors, and vague.

 

 

Author Response

Reviewer: 2

1. Short Summary: 

(1) explain all abbreviations the first time they are used

spellcheck including punctuations and blanks.

Response:

Thank you very much for raising this issue, we have already added the full names of all abbreviations. What’s more, we have checked the spelling of our paper.

(2) table 1 pops out of the blue, why is it where it is and what does it tell us?

the algorithms need to be presented in tables or figures and referred to, instead of pseudo code I would appreciate a flowchart

where does table 2 belong to?

Response:

Thank you very much for raising this issue, we have referred table 1 in section 2.1.

Due to the limitation of paper length, we add the flowcharts in our response letter and show them as follows.

(3) how is data management considered in the model presented?

to assess the relevance of the simulxation and findings: can you give examples for the different tasks and relate to savings

Response:

Thank you very much for raising this problem. This paper supposes that data management takes into account the data size (size) and calculation amount (length) in the model, which will affect the transmission delay and calculation delay. Different types of tasks can be considered to be computationally intensive and IO-intensive, which are reflected in the case where the data length is much larger than the size or vice versa. For computationally intensive applications, there is a higher power saving, because computationally intensive applications have large differences in power consumption due to different computing locations, and for IO-intensive task models, they will try to place them in edge node computing. The space for the algorithm to be schedulable is small.

(4) The conclusions are brief, contain several language errors, and vague.

Response:

Thank you very much for raising this problem. We have revised the conclusion section as follows.

 

 

Author Response File: Author Response.docx

Reviewer 3 Report

 The paper utilizes edge computing to reduce latency and power consumption. Typically, the authors aim to solve the resource allocation problem of edge servers with constrained resources in edge computing. They focus on energy efficiency for multiple users via edge devices with reduced power while meeting delay and resource constraints. They then propose 2 kinds of approaches: scheduling scheme by offloading task to the edge nodes and resource provisioning strategy by splitting a node into two subnodes (slave and master). The proposal is reasonable, however, there are lots of existing issues that the authors need to address for the possible publication of the research, as follows:

 

Regarding the paper content:

• The reviewer’s first concern is that this manuscript is an extended version of the authors’ paper presented at INFOCOM workshop, but it is never referred to in the text (as a reference). Also, the authors are expected to explicitly discuss the differences between this paper and their mentioned prior workshop paper at the end of the Introduction Section. Please also mention the title of the INFOCOM paper on the first page. Please revise the paper accordingly.
• The paper includes Mobile edge computing (MEC) in the title, but most of the content is about edge computing only. Please do either revise the paper title or clarify MEC and how the authors applied MEC in the proposal.
• Please add the full name of a term when it is first mentioned in the text, e.g., MIPS.
• In the Related work section, the authors should add more detail on how the referred papers address the Energy Efficiency problem using Edge computing. Only some of them include the relevant overall work on this topic in the current version of the manuscript.
• Section titles are wrong (the paper structure paragraph right before the Related work section). The paper should have the first section entitled “Introduction”, followed by “Related work” as section 2, and so on.
• 3 should be changed into Fig. 2 and vice versa since it is introduced before the current Fig. 3
• The task and mobile index should be denoted differently (both are currently denoted by i)
• The overall writing is Ok, but there are some existing writing errors in the manuscript (typically, grammatical errors and incomplete sentences), e.g., grammar errors at the first paragraph in the Introduction section Please revise accordingly.

 

Regarding the problem Formulation:

- Please explain N_0 and omega in Equation (3).

- The channel power gain expression should be corrected: d_ij instead of d_i.

 

Regarding the Scheduling algorithm:

-  Please analyze the complexity of the heuristic approach in Algorithm 2 (SSBM).

- Please add a reference for this statement” For a normal PC, this power consumption is about 4.5W.”

- Please explain why each node of the tree should have 3 children.

 

Regarding the Resource provisioning algorithm:

- The referred line number for Algorithm 3 to describe the proposed resource provisioning strategy is incorrect. Please revise it accordingly.

 

Regarding the evaluation section:

• Network topology for scheduling evaluation is relatively simple in which only 5 edge nodes are used, and 5 users connecting to each edge node. The authors are expected to perform larger-scale experiments to show the scalability of the proposal.
• What is the Coefficient of power consumption alpha? Why it is set as 10? Also, why the penalty coefficient is 630?
• To verify the authors’ statement regarding the total delay of the mobiles, please add another (sub)Fig. where the number of executions is less than 5,000.

 

Minor issues:

• Trinh, Trinh et al. [20], Gupta Trinh et al. [24] (should be Gupta only).
• D_i^t is not in equation (2).
• What is “Section 4 of Section 3”?
• 3 : the emit emit interval.

Author Response

Reviewer: 3

1. Short Summary: 

The paper utilizes edge computing to reduce latency and power consumption. Typically, the authors aim to solve the resource allocation problem of edge servers with constrained resources in edge computing. They focus on energy efficiency for multiple users via edge devices with reduced power while meeting delay and resource constraints. They then propose 2 kinds of approaches: scheduling scheme by offloading task to the edge nodes and resource provisioning strategy by splitting a node into two subnodes (slave and master).

Response summary:

We would like to thank the reviewer for carefully reviewing our paper. Many constructive comments are offered. We have revised our paper according to your suggestions. We provide point-to-point responses to each of your comments.

(1) The reviewer’s first concern is that this manuscript is an extended version of the authors’ paper presented at INFOCOM workshop, but it is never referred to in the text (as a reference). Also, the authors are expected to explicitly discuss the differences between this paper and their mentioned prior workshop paper at the end of the Introduction Section. Please also mention the title of the INFOCOM paper on the first page. Please revise the paper accordingly.

Response:

Thank you very much for raising this problem. This paper is an extended version of the INFOCOM workshop paper, the difference was concluded in our cover letter. The main content of our cover letter list as follows.

A preliminary version of this work has been accepted by IEEE INFOCOM 2020. We have included substantial extension over the preliminary version and a summary of differences is included here for your reference.

• We first extend our problem into multiple users and propose a two stages scheduling strategy which includes the scheduling and resource provisioning.

The published version by IEEE INFOCOM 2020 mainly introduces the resource provisioning stage. While this paper focuses on the scheduling problem between users and edge nodes with energy efficiency. In Section 2, we re-formulate the problem and re-define the model. Base on that, we propose two energy-efficient scheduling strategies in Section 3.

• We propose an efficient scheme based on the branch and bound method for the scheduling stage and discuss the complexity in Section 3.1.

In Section 3.1, we first propose a scheduling scheme based on the branch and bound method in Algorithm 1, which is a fast search tree algorithm based on a depth-first searching algorithm. Then we discuss the complexity of this algorithm which is .

• Furthermore, we propose a heuristic algorithm that minimizes the power consumption and guarantees the users' deadline as well as lower complexity in Section 3.2.

For Algorithm 1, the time complexity increases exponentially with the expansion of sub-tasks. Thus, we propose a heuristic algorithm in Algorithm 2. The main idea of the algorithm is to select the best node in every layer. The detailed description of this algorithm is shown in Section 3.2.

• We conduct various simulations for our scheduling strategy by using the platform we designed. The results are shown from different perspectives to provide conclusions in Section 5.1.

(2) The paper includes Mobile edge computing (MEC) in the title, but most of the content is about edge computing only. Please do either revise the paper title or clarify MEC and how the authors applied MEC in the proposal.

Response:

Thank you very much for raising this problem. In this paper, we focus on reducing the power consumption in edge computing by considering the resources and delay constraints. In order to avoid the confusion, we have revise the paper into edge computing.

(3) Please add the full name of a term when it is first mentioned in the text, e.g., MIPS.

Response:

Thank you very much for raising this issue, we have already added the full names of all abbreviations.

(4) In the Related work section, the authors should add more detail on how the referred papers address the Energy Efficiency problem using Edge computing. Only some of them include the relevant overall work on this topic in the current version of the manuscript.

Response:

Thank you very much for raising this issue, we have already added several references in section 2.

(5) Section titles are wrong (the paper structure paragraph right before the Related work section). The paper should have the first section entitled “Introduction”, followed by “Related work” as section 2, and so on.

Response:

Thank you very much for raising this issue, we have already revised it as follows.

 

(6) 3 should be changed into Fig. 2 and vice versa since it is introduced before the current Fig. 3

The task and mobile index should be denoted differently (both are currently denoted by i)

Response:

Thank you very much for raising this issue, and we have revised the label and position of this figure. What’s more, the index of the task has been revised into k.

 

(7) The overall writing is Ok, but there are some existing writing errors in the manuscript (typically, grammatical errors and incomplete sentences), e.g., grammar errors at the first paragraph in the Introduction section Please revise accordingly.

Response:

Thank you for offering this recommendation. We have revised the grammar errors in our paper.

(8) Regarding the problem Formulation:

- Please explain N_0 and omega in Equation (3).

- The channel power gain expression should be corrected: d_ij instead of d_i.

Response:

Thank you very much for raising this issue. In our paper, N_0 denotes the channel white Guassian noise, and omega denotes the channel path loss exponent. We have added the description in lines 133 to 138.

(9) Regarding the Scheduling algorithm:

- Please analyze the complexity of the heuristic approach in Algorithm 2 (SSBM).

- Please add a reference for this statement” For a normal PC, this power consumption is about 4.5W.”

- Please explain why each node of the tree should have 3 children.

Response:

Thank you for offering this recommendation. We analyze the complexity of the heuristic approach in Algorithm 2 (SSBM), which has the worst time complexity of SSBM is O(n^3).

We have added one reference for this statement in line 241.

In this paper, we suppose that each node of the tree has 3 children. This assumption can be extended to any other number which is larger than 1.

(10) Regarding the Resource provisioning algorithm:

- The referred line number for Algorithm 3 to describe the proposed resource provisioning strategy is incorrect. Please revise it accordingly.

Response:

Thank you very much for raising this issue. We have already revised it as follows.

 

(11) Regarding the evaluation section:

Network topology for scheduling evaluation is relatively simple in which only 5 edge nodes are used, and 5 users connecting to each edge node. The authors are expected to perform larger- scale experiments to show the scalability of the proposal.

What is the Coefficient of power consumption alpha? Why it is set as 10? Also, why the penalty coefficient is 630?

To verify the authors’ statement regarding the total delay of the mobiles, please add another (sub)Fig. where the number of executions is less than 5,000.

Response:

Thank you very much for raising this issue. In our simulation, we use 5 edge nodes and 5 users which can also extend to a larger number. Since the tendency of the results are nearly the same, so we choose 5 edge nodes and 5 users to show the experimental results. In our experiment, there exist capacity limitation of our machines, so that, it is hard to do the larger-scale experiments. In our future, we try to expand the scale of our experiments by renting high-performance servers.

The settings of alpha and the penalty coefficient are referred in the references [25]-[29], which have been mentioned in lines 129 and 130.

(12) Minor issues:

Trinh, Trinh et al. [20], Gupta Trinh et al. [24] (should be Gupta only). D_i^t is not in equation (2).

What is “Section 4 of Section 3”?

3 : the emit emit interval.

Response:

Thank you for offering this recommendation. We have revised these problems in our paper.

 

Author Response File: Author Response.docx

Round 2

Reviewer 3 Report

The paper utilizes edge computing to reduce latency and power consumption. Typically, the authors aim to solve the resource allocation problem of edge servers with constrained resources in edge computing. They focus on energy efficiency for multiple users via edge devices with reduced power while meeting delay and resource constraints. They then propose 2 kinds of approaches: scheduling scheme by offloading task to the edge nodes and resource provisioning strategy by splitting a node into two subnodes (slave and master). The proposal is reasonable in the context of energy-efficient resource provisioning using Edge computing, and in the revised manuscript, the authors addressed most of the reviewer’s concerns. However, there are a few existing issues that the authors need to address for the possible publication of the research, as follows:

- First and foremost, the INFOCOM paper title needs to be mentioned by the abstract so that the readers can find its content if needed. Also, the authors need to explicitly state the difference between the proposal and the mentioned INFOCOM workshop paper (as provided in the response letter to the reviewer) in the Introduction of the paper. Besides, please add the INFOCOM paper and refer to it accordingly in the manuscript. All of these need to be reflected in the final manuscript.

- Regarding the Related work section, even the authors added some new related studies, they need to briefly analyze how the referred (existing) papers address the Energy Efficiency problem using Edge computing explicitly. These expected analyses are also not yet presented in the revised manuscript.

- Regarding the evaluation section, please explain why you mentioned this: “Since the tendency of the results are nearly the same, so we choose 5 edge nodes and 5 users to show the experimental results.” How do the authors state that the tendency of the results is the same while the scope of the experimental evaluation is very limited?

Author Response

1. Short Summary: 

The paper utilizes edge computing to reduce latency and power consumption. Typically, the authors aim to solve the resource allocation problem of edge servers with constrained resources in edge computing. They focus on energy efficiency for multiple users via edge devices with reduced power while meeting delay and resource constraints. They then propose 2 kinds of approaches: scheduling scheme by offloading task to the edge nodes and resource provisioning strategy by splitting a node into two subnodes (slave and master). The proposal is reasonable in the context of energy-efficient resource provisioning using Edge computing, and in the revised manuscript, the authors addressed most of the reviewer’s concerns. However, there are a few existing issues that the authors need to address for the possible publication of the research, as follows:

Response summary:

We would like to thank the reviewer for carefully reviewing our paper. Many constructive comments are offered. We have revised our paper according to your suggestions. We provide point-to-point responses to each of your comments.

(1) First and foremost, the INFOCOM paper title needs to be mentioned by the abstract so that the readers can find its content if needed. Also, the authors need to explicitly state the difference between the proposal and the mentioned INFOCOM workshop paper (as provided in the response letter to the reviewer) in the Introduction of the paper. Besides, please add the INFOCOM paper and refer to it accordingly in the manuscript. All of these need to be reflected in the final manuscript.

Response:

Thank you very much for raising this problem. According to the template provided by the Applied Science Journal, we have added the summary in our manuscript as follows.

The difference between this manuscript and the conference version is mentioned in Section 1 in lines 43 to 57.

(2) Regarding the Related work section, even the authors added some new related studies, they need to briefly analyze how the referred (existing) papers address the Energy Efficiency problem using Edge computing explicitly. These expected analyses are also not yet presented in the revised manuscript.

Response:

Thank you very much for raising this problem. We have revised Section 2 in Line 79 to 122 as follows.

(3) Regarding the evaluation section, please explain why you mentioned this: “Since the tendency of the results are nearly the same, so we choose 5 edge nodes and 5 users to show the experimental results.” How do the authors state that the tendency of the results is the same while the scope of the experimental evaluation is very limited?

Response:

Thank you very much for raising this issue. In our simulation, we use 5 edge nodes and 5 users which can also extend to a larger number. In order to show the tendency, we do the simulation by setting the following metrics.

As shown in the above figures, the tendencies of the results are nearly the same. What’s more, when the users and edge nodes are small (Group 1 to Group 3), the results are not obvious. When the numbers of users and edge nodes become large, the gaps between GAB and the other three algorithms are too large, which leads to insignificant differences in results between some algorithms. Thus, we choose the median (5 edge nodes and 5 users) to show the experimental results.

Author Response File: Author Response.docx