A Novel Adaptive East–West Interface for a Heterogeneous and Distributed SDN Network
Round 1
Reviewer 1 Report
The paper proposes a East-West Interface (called SINA) for a Heterogeneous and Distributed SDN Network. This interface leads to different SDN controllers to exchange status information. The clear description of the implemented libraries is done, and also it is describet the way to introduce them in three widely-used SDN controllers. Authors further propose a RL-based consistency algorithm to overcome the problem of scalability but keeping a consistency solution. The performance section is divided in two phases: in the first one, authors valided the proposed interface, while in the second one, author tested the performance of an adaptative Quorum-based replication mechanism.
The topic is very interesting and timeliness; moreover, the paper is well-written and organized. The quality of the presentation is good. The proposed schemes and algorithms are explained weel, although it needs to be improved for some weak points. For instance, to improve the readability, I suggesto to move Table 1 at the end of Section 2.
The definition of read-quorum is a little bit tricky to understand. Then, the use of the same letter to define different parameters is not useful. For instance, r and R are used about rewards and requests.
About testbed setup for the validation phase, it is clear that there is one virtual machine dedicated for each SDN controller (up to 12). Are the links between them also implemented in mininet? And are there dedicated links or do they follow the network topology under test? I think this is an important note to understand the latency metric.
Few typos need to be checked, like:
row 142: HyperFlow instead of HyperFlox
row 216: "@Active" has to be written in italic
Author Response
Please see the attachment.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
I appreciate the authors hard work on this paper. In my opinion, the distributed architecture in SDN and consistency between controllers are the important topic in this area.
I suggest several comments to improve the readability and quality of this paper.
- A natural question is that since the paper claims the two approaches in this paper both perform well (in Section I), which method is more suitable for a specific SDN system?
- One possible consideration with using multiple SDN controllers is to take virtualizations. For example, several network hypervisors for SDN (such as in the following papers) host multiple controllers and communicate with each controller in a centralized way. Could you discuss the possibility of this direction and its problems?
- Jin, Xin, et al. "CoVisor: A Compositional Hypervisor for Software-Defined Networks." 12th USENIX Symposium on Networked Systems Design and Implementation (NSDI 15). 2015.
- Yang, Gyeongsik, et al. "Libera for programmable network virtualization." IEEE Communications Magazine 58.4 (2020): 38-44.
- Between the heterogeneous SDN controllers, the form of the network state stored in each controller might differ, which requires a sort of translation. Does the SINA in the application plane (Fig. 2) have such functionalities?
- Is there any measurement to justify the necessity of an RL-based consistency algorithm? Such a measurement could be the numbers for the network bandwidth and the QoS degradation from the broadcast.
- Consistency within distributed systems is a well-investigated area as a name of "consensus." Could you compare (or evaluate) your mechanism with the consensus algorithms, such as Kafka and Raft?
- The descriptions presented below Table 2 that explain SDN controller compositions are difficult to recognize. Moreover, the table's first column is the network but it in fact has the SDN controllers.
- What does the overhead refer to on the left y-axis of Fig. 6? Overhead is such a general term that it cannot be understood accurately.
- The evaluations (e.g., Fig. 6) compare the overheads with each SDN controller. A better comparison could include the previous studies, such as DSF, CIDC, and CSM, which the authors summarized.
- Additionally, for the experiments regarding the consistency, the results could be interpreted much better using the comparisons with other consistency mechanisms.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Round 2
Reviewer 2 Report
I appreciate the authors' efforts in replying to the comments.
Although my curiosity has been entirely resolved, most aspects of my concerns have been addressed and solved.
I respectfully recommend that the authors reflect the answers to the previous reviews on the paper (e.g., discussion or future work) if it is possible.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
