Software-Defined Heterogeneous Edge Computing Network Resource Scheduling Based on Reinforcement Learning
Round 1
Reviewer 1 Report
This paper designs a software defined heterogeneous sensor network architecture. In this paper, task scheduling is assigned to edge nodes for processing. With energy consumption and energy balance as the optimization objectives, resource scheduling problem is modeled as a Markov decision process, and an approximate strategy optimization resource scheduling algorithm is designed to achieve the optimization objectives. The structure of this paper is complete and the idea is clear, and the reviewer would suggest some further modifications as follows:
1. The paper gives the expression of the optimization problem, please add the corresponding constraints and explain the constraints.
2. What is the process of task allocation and scheduling ? Section 2 describes that the computing task is transmitted from the remote data center to the edge server, the edge server is assigned to the edge computing node to complete the computing task and finally returns to the data center, but this seems inconsistent with the procedure shown in Figure 2.
3. Some symbols in formula (6) are not explained in the paper, please add corresponding notes. Is it the greater the energy balance, the better the performance?
4. Please briefly explain why energy balance in Figure 5 increases first and then decreases with the number of tasks.
Author Response
Please kindly find our response at the attachment.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
Comment1：
The model and idea of the whole article are very good, which is a good article. The dotted line in Figure 2. coincides with the text. It is recommended to change the two lines to not overlap. The layout and format of the paper need to be further optimized. Overhaul is recommended.
Comment2：
What impact does different types of data have on the model, does matter? It can be explained in more detail. The advantages of the separation of control layer and data layer can be further explained.
Comment3：
and needs to explain its meaning in detail. What is the impact of the transmission distance threshold d0 on the model and optimization target? Etotla will explain the meaning of total energy consumption by the way, and all the variables appearing for the first time will be marked.
Comment4：
Network load balancing can further explain the physical meaning, and add some content in the text to explain it. λ Is there a value range? The formula for PPO can be explained in detail. Add more formulas.
Author Response
Please kindly find our response at the attachment
Author Response File: Author Response.pdf
Reviewer 3 Report
This work considers a heterogeneous edge computing network with heterogeneous edge computing nodes and task requirements. We design a software-defined heterogeneous edge computing network architecture to separate the control layer and the data layer. The mathematical derivation is quite sufficient, and the logic is meticulous. But the work still has the following problems:
1. The network energy consumption and load balance are analyzed in detail in the simulation analysis section. We observe that the network equilibrium exhibits irregular jitter with the number of tasks. The reasons for this change suggest further elaboration by the author.
2. The authors use reinforcement learning to solve the resource scheduling problem, which is very innovative. However, whether reinforcement learning can achieve the optimal solution is not discussed in detail. We suggest that the authors elaborate further on the optimality analysis.
3. What are the contraints for problem (6)?
4. The authors may think about more state-of-the-art DRL based edge computing work, such as, edge intelligence for energy-efficient computation offloading and resource allocation in 5G beyond.
Author Response
Please kindly find our response at the attachment
Author Response File: Author Response.pdf
