An Approach Based on Knowledge-Defined Networking for Identifying Heavy-Hitter Flows in Data Center Networks

Round 1

Reviewer 1 Report

The paper presents an example to apply Machine Learning techniques to Data Center Networks. Knowledge Defined Networking architecture is recommended to detect heavy-hitter flows. This problem has large practical relevance because HH-flows are crucial from the aspects of the performance of the system. The architecture consists of data acquisition, analyser and application modules to route flows efficiently. The novelty of the system is represented by applying clustering techniques to identify heavy-hitter flows instead of the current threshold-based techniques.

The paper is carefully written. The paper provides appropriate overview on the background and the preliminaries. The proposed architecture is clearly presented. The method is evaluated on publicly accessible traffic trace. The proposed method reduces the number of rules applied on network devices.

My main concern is that the paper does not perform evaluation to examine whether applying the proposed approach really improves the overall network performance.

There are various methods to determine the number of clusters. It would be interesting to try other ones as well besides the Silhouette method.

Some further comments:

line 63: routes -> route

line 75: KDN-based -> KDN

 

Author Response

RE: Response Letter

An Approach based on Knowledge-Defined Networking for Identifying Heavy-Hitter Flows
in Data Center Networks
Manuscript ID: applsci-620554

Dear Reviewer,

Thank you for your review of our paper, and for the comments and suggestions that ensued. A major revision of the paper has been carried out to take all of them into account.

In the present response letter, we first detail the major changes that have been made in
the paper to correct the main weaknesses identified by the review. Then we sequentially
address all of the points raised.

To facilitate reading, we have highlighted all corresponding changes in blue.

Thank you very much.

Sincerely,

Alejandra Duque-Torres
Eng. Electronica - Univ. del Quindio
M.Sc (C) Telematic Engineering -
Universidad del Cauca

Author Response File: Author Response.pdf

Reviewer 2 Report

The paper presents a method for classification of HH and non-HH flows based on a ML approach.
The first observation concerns the applicability of the proposed methed. The study is accurate and the obtained results (i.e. the values of the thresholds of flow size and number of packets) seems significant. However, the main limit of the proposed classification method is that, as the authors themselves underline, it requires a large time to be completed.
A study of this particular aspect is not carried out, however it is absolutely indispensable to the aim of understanding the usability of the method in real scenarios.
Moreover, the results of the proposed method should be compared with those deriving from other approaches.

A second observation is that that the authors propose the use of MiceDCER algorithm to route the non-HHs flows. Though a reference is given, in the paper this algorithm is not discussed for the specific application of DCN considered. This lack of details makes impossible to evaluate the soundness of the results shown in fig. 6.

In summary, the above mentioned remarks rise several doubts on the actual utilization of the proposed method for application in data Center.

A further remark is that a great percentage of the paper is devoted to tutorial aspects. These parts could be reduced.

Author Response

RE: Response Letter

An Approach based on Knowledge-Defined Networking for Identifying Heavy-Hitter Flows
in Data Center Networks
Manuscript ID: applsci-620554

Dear Reviewer,

Thank you for your review of our paper, and for the comments and suggestions that ensued. A major revision of the paper has been carried out to take all of them into account.

In the present response letter, we first detail the major changes that have been made in
the paper to correct the main weaknesses identified by the review. Then we sequentially
address all of the points raised.

To facilitate reading, we have highlighted all corresponding changes in blue.

Thank you very much.

Sincerely,

Alejandra Duque-Torres
Eng. Electronica - Univ. del Quindio
M.Sc (C) Telematic Engineering -
Universidad del Cauca

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

I accept the answers of the authors. I recommend accepting the paper.

Reviewer 2 Report

I appreciated the efforts of the authors to meet the observations of the reviewers.

I thiink that the paper is really improved.

Anyway, the doubts about the actual implementability of the method in data center have not been completely eliminated.

A performance study of the complete system (flow identifications + routing) would be necessary to understand the advantage on the performance of a Data Center.

However, since the above study is really time consuming and as the paper containg somo interesting results, I think that it canbe published on Applied Sciences.