A DRL-Based Task Offloading Scheme for Server Decision-Making in Multi-Access Edge Computing
Round 1
Reviewer 1 Report
The authors address a relevant and timely topic of Multi-Access Edge Computing (MEC), which is crucial for supporting the next-generation Internet of Things (IoT) networks. The adoption of a deep reinforcement learning-based Soft Actor-Critic (SAC) approach showcases the use of advanced methodologies to address challenges in MEC environments and this approach indicates the paper's sophistication and innovation. The paper goes beyond oversimplified scenarios by considering a multi-user multi-MEC server environment and this realistic approach acknowledges the complexities of real-world MEC settings. The proposed SAC-based approach aims to reduce latency and improve energy efficiency, two critical factors in enhancing the performance of MEC systems and this contribution aligns with the goals of optimizing resource allocation. The paper conducts experimental evaluations that demonstrate the rapid and stable convergence of the proposed algorithm and these results substantiate the effectiveness of the approach and its superiority over existing methods.
The authors are highly encouraged to consider the following suggestions provided by the reviewer in order to further improve the scientific depth and the quality of presentation of the manuscript. The paper mentions several open challenges in MEC environments, however, it does not provide a clear and detailed problem statement that outlines these challenges comprehensively. The paper briefly mentions the use of the Soft Actor-Critic (SAC) algorithm without providing a thorough explanation of how it works, its adoption in previous research works, e.g., Price and Risk Awareness for Data Offloading Decision-Making in Edge Computing Systems, doi: 10.1109/JSYST.2022.3188997, its advantages, and how it's adapted to the specific context of MEC. The paper claims superior performance over existing methods, however, it lacks a direct comparative analysis that showcases the performance metrics and quantifies the improvements achieved. The paper discusses the modeling of dependent task division problems within the Markov Decision Process (MDP), however it doesn't provide enough clarity on how the state space, action space, and reward function are defined and utilized. The paper doesn't discuss how the proposed approach can be practically applied in real-world scenarios, the potential challenges in implementation, or any practical limitations. The manuscript needs a major revision to address the above comments.
Minor editing of English language required
Author Response
Please see the attachment.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
This study investigates dependent task offload scheduling, with the goal of optimizing both latency and energy consumption in multi-server and multi-SD installations. The authors combined a collaborative architecture between MECs, framing it in the context of MDP, and proposed SACTOS, which operates under centralized control, significantly reducing service latency and conserving energy. The authors claim that this method outperforms the other existing methodologies.
There are certain queries.
1. How has the reward function been taken? What is the physical basis? Explain clearly the steps.
2. Comparisons are made regarding latency and energy conservation, what about bandwidth limitations?
3. Can the simulated results be verified using a prototype? What is the cost of deployment and maintenance for a realistic system?
4. Can the authors compare different methodologies w.r.t. storage capacity and security?
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 3 Report
The work focuses on a deep reinforcement learning-based Soft Actor-Critic (SAC) approach to utilize MEC server decision-making to handle tasks in a multi-user, multi-MEC server environment. The problem being addressed is well formulated, and the solution method is adequately presented. The work poses good significance, and the results are quite interesting. However, I still have some major concerns that need to be addressed.
1. The authors stated that the proposed approach, which employs SAC for computing offloading and MEC server decision-making, is demonstrated via numerical results to significantly reduce latency and improve energy efficiency. It is not convincing why the SAC was used in this study, given the fact several other related works have employed it extensively.
2. In Table 1, the authors presented a comparison of other approaches. I wanted to see how the projected method compared to other existing methods but this is missing in the Table. Please add your method to the Table.
3. In the system architecture shown in Figure 1, several useful information regarding the interconnections are missing. The authors must redraw the architecture to show the various links and interfaces and their significance.
4. It would be nice to provide the references to the equations. Most of the equations require their full proofs. Please provide more useful information relevant to understanding the equations.
5. The problem statement ought to come before the system architecture. Please revise the structure of the manuscript for good readability.
6. It is unclear why the authors formulated the offloading problem as a Markov Decision Process. Please provide adequate clarity on this choice.
7. The structure of the SAC scheme in Figure 2 is lacking in description and explanation.
8. The axes of Figures 3-7 need to be revised. The font size is too small. Please revise all accordingly.
9. Further discussions on Figures 5 and 6 are required. The trend and growth pattern requires an elaborate explanation.
10. At least three additional related works from 2023 need to be added and discussed appropriately.
Minor English editing is required.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Round 2
Reviewer 1 Report
The authors have addressed in detail the reviewers comments.
Reviewer 2 Report
Can be accepted
Reviewer 3 Report
The authors made an attempt to address my earlier comments.
The English reporting is good. However, the paper will benefit from minor editing.
