1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
|
专家1：
审核意见：
This paper addresses a critical challenge in Mobile Edge Computing (MEC), which is the efficient placement of services to minimize delay and migration costs in a dynamic, user-mobility-aware environment. The methodology presented in the paper is innovative and well-founded. However, there are still some issues that need to be improved.
(1) The paper is well-written. However, there are a few grammatical errors and awkward phrases that need attention. For example: "We leverages the service priority to minimize the delay of delay-sensitive applications..." should be "We leverage service priority to minimize the delay for delay-sensitive applications...", and "We propose a algorithm with service priority to solve the delay-sensitive service placement problem." should be "We propose an algorithm with service priority to solve the delay-sensitive service placement problem."
(2) It would be beneficial to include more diverse scenarios in the experiments to demonstrate the robustness of the proposed algorithm under different conditions.
(3) While the methodology is well-conceived, providing a flowchart of the DRL algorithm can help readers understand the implementation better.
(4) Consider including more evaluation metrics beyond delay and migration costs, like computational overhead, could provide a more comprehensive assessment of the proposed method.
专家2：
审核意见：
This paper proposes an architecture to reduce the delay of delay-sensitive applications in the cloud-edge-device scenario, and uses an active service placement method based on reinforcement learning to avoid frequent service migration caused by user movement, and the service will migrate to the expected arrival position of users with multi-step trajectory prediction. This paper also uses real data sets to verify the proposed method in a simulated environment, which can effectively reduce the migration cost and delay of sensitive applications.
1. In the related work, the service placement problem in MEC is discussed from two aspects of existing research, and the limitations and shortcomings of each method are pointed out by adding critical analysis, and how these limitations guide the future research direction is discussed. For each method, it provides more in-depth implementation steps, technical challenges and real application examples to help readers better understand their advantages and disadvantages.
2. The expression of Unuse in Figure 1 is unclear, and it is impossible to see what role it plays in this scene in the figure, and this part is not explained in the text.
3. The processing delay of edge server and cloud server is clearly described, and the process of request processing and task migration is also well described. In the experimental part, the effectiveness of reducing service delay can be clearly seen.
4. In this paper, PriMSP is compared with DRL-SP, Pri-SP and T-SP. Although this paper shows that the existing method is superior to the other three methods, it does not deeply study the potential limitations of the proposed algorithm. Discussing these limitations and outlining the direction of future work will strengthen the arguments put forward in this paper.
专家3：
审核意见：
This paper proposed a mobility-aware service placement strategy in the cloud-edge-device scenario to reduce the average delay of delay-sensitive applications. A DRL-based active service placement algorithms combining service priority and trajectory prediction is provided. The authors demonstrate the algorithm can reduce delay-sensitive applications delay via a series of experiments. The paper is well organized, but there are still some concerns.
1. In the System Model of Section3.1, For the ratio term d(u_i, m_k)/λ_i and d(m_k, m_p)/λ_i, what are the units of d(x, y). and what are the units of λ_i? Can these ratio terms represent delay? There may be some problems with System Model. Are the datasets used in the Experiments relevant to System Model?
2. There is a confusion of notations in the paper. Is p_i the transmission power or the service priority?
3. Some notations should be interpreted at the first occurrence, e.g. r_i, o_i and PriMSP.
4. Fig. 2 should include the overall comparison of all algorithms.
5. The migrations times in Fig. 3 should be labeled with units.
|