Containerization in Edge Intelligence: A Review
Round 1
Reviewer 1 Report
Comments and Suggestions for AuthorsThe authors explained the advantage of container technology in software development for edge intelligence. They tried to solve the problems caused by the distance between data source and cloud such as heterogeneity of devices used in edge computing. In this paper, the authors surveyed various issues and development results related to containerization in edge intelligence.
In order for this paper to be published in the Electronics journal, the authors need to add, correct, and supplement the following issues.
(1) In the containerization in edge intelligence, the processing of the data itself is important. So, the authors should add an explanation of containerization from a data perspective.
(2) In order to illustrate the practical applicability of the descriptions presented in this paper, it is necessary to add a simple case study using machine learning approach.
(3) The authors explain existing research on the containerization in edge intelligence in sections 2 to 5 for each topic. In addition, if content that can summarize and explain the entire process for the containerization in edge intelligence is added, the understanding of the paper can be improved.
Author Response
Dear reviewer, you can find answers to your comments in the points below. We tried to process all your comments and adjust our paper according to the instructions. Our answers to the comments are available below, also describing the necessary changes which we made within the manuscript. The changes which we made in the manuscript are also displayed in the attached Difference.pdf, where the blue text represents new additions and the olive text that was moved.
(1) In the containerization in edge intelligence, the processing of the data itself is important. So, the authors should add an explanation of containerization from a data perspective.
We have added a new subsection, 3.4 entitled “Containers and Data” where we briefly describe the statelessness of containers and the use of volumes, as well as the typical approach to inserting data into containers.
(2) In order to illustrate the practical applicability of the descriptions presented in this paper, it is necessary to add a simple case study using machine learning approach.
We are currently working on such a study, the base of which is available in a published paper available at https://www.mdpi.com/1424-8220/23/17/7662. We are currently extending this framework to include machine learning and artificial intelligence methods, the results of which will be published later.
(3) The authors explain existing research on the containerization in edge intelligence in sections 2 to 5 for each topic. In addition, if content that can summarize and explain the entire process for the containerization in edge intelligence is added, the understanding of the paper can be improved.
We have added a new subsection, 3.8 entitled “Creating a Container Image”, which contains an example and a description of how to containerize a Python application.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
Comments and Suggestions for Authors
This paper reviews the containerization in Edge Intelligence.
1. There have a misunderstanding in the introduction "[4] mention that the amount of data generated per person was around 1.7 MB in 2022, which came out to around 44 zettabytes per day in total."(line 16) The corresponding original paper of [4] means "every Humans generate 1.7MB of data every second.”
2. In Section 3.1, for the two typical hardware virtualization hypervisors, Type 1 Hypervisor and Type 2 Hypervisor, what are the specific performance differences between them, such as effective CPU utilization, memory usage, I/O memory access, etc. Please make a brief addition.
3. The division of Chapter 3 is confusing. It is recommended that the introduction of operating system-level virtualization performance in Section 3.5 be included in Section 3.2.
4. In addition to Docker and k8s, Anaconda, as a commonly used environment container, also occupies a certain market share in academic research and production applications involving the field of edge intelligence. You can briefly introduce the characteristics of Anaconda, point out its differences and connections with containers such as Docker, and prove that Anaconda does not belong to a container in a strict sense. That increases the persuasiveness of the article.
5. For the description of the levels of edge intelligence in Chapter 5, it is recommended to use pictures or tables to show the training and inference of the device, edge and cloud in the six levels.
Comments on the Quality of English LanguageSome sentences in the article are lengthy or unclear, such as “The transition from the cloud environment, made up of large and powerful servers to an edge environment with small devices offering only a fraction of the performance of the cloud alternatives poses a large set of challenges and opportunities [8–10].”(line 52) and “The Cognitive Speech Service container can be deployed on the edge or in the cloud if sufficient performance is not available on the edge.”(line 594), etc.
Author Response
Dear reviewer, you can find answers to your comments in the points below. We tried to process all your comments and adjust our paper according to the instructions. Our answers to the comments are available below, also describing the necessary changes which we made within the manuscript. The changes which we made in the manuscript are also displayed in the attached Difference.pdf, where the blue text represents new additions and the olive text that was moved.
- There have a misunderstanding in the introduction "[4] mention that the amount of data generated per person was around 1.7 MB in 2022, which came out to around 44 zettabytes per day in total."(line 16) The corresponding original paper of [4] means "every Humans generate 1.7MB of data every second.”
We have fixed the sentence, it now accurately says “data generated per person was around 1.7 MB per second in 2022”
- In Section 3.1, for the two typical hardware virtualization hypervisors, Type 1 Hypervisor and Type 2 Hypervisor, what are the specific performance differences between them, such as effective CPU utilization, memory usage, I/O memory access, etc. Please make a brief addition.
We have added a new subsubsection, 3.1.3 which describes the test results in more detail.
- The division of Chapter 3 is confusing. It is recommended that the introduction of operating system-level virtualization performance in Section 3.5 be included in Section 3.2.
Chapter 3.5 was moved above, it is now 3.3.
- In addition to Docker and k8s, Anaconda, as a commonly used environment container, also occupies a certain market share in academic research and production applications involving the field of edge intelligence. You can briefly introduce the characteristics of Anaconda, point out its differences and connections with containers such as Docker, and prove that Anaconda does not belong to a container in a strict sense. That increases the persuasiveness of the article.
The subsection explaining Anaconda and more specifically its main feature conda was added as subsection 4.3 with a brief description of the features it offers and how it differs from containers.
- For the description of the levels of edge intelligence in Chapter 5, it is recommended to use pictures or tables to show the training and inference of the device, edge and cloud in the six levels.
We have added Table 1, present at page 12, to summarize where and how the training and inference take place.
Author Response File: Author Response.pdf
Round 2
Reviewer 2 Report
Comments and Suggestions for AuthorsIn Chapter 5, for the description of the six levels of edge intelligence, it is suggested that the brief descriptions of each level in line 451- 464 should be distributed to the subsequent detailed descriptions of each level (line 477 - 559). So as to avoid duplication with Table 1.
