Hop-by-Hop Multipath Overlay Routing for Optimizing Network Resource Allocation in WANs

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

Dear authors, in the next paragraphs my comments about your manuscript.

 

This article analyzes and proposes a multipath routing approach based on intermediate nodes, meeting an interesting distinction from the classical source-based approach, thereby creating a more dynamic response against network congestion.

The paper presents an analysis using multiple real topologies (TA2, Germany50, Abovenet) and an artificially generated topology. Results remained consistent, thus strengthening the argument for the usefulness of the proposed algorithms.

The authors show that: MOP increases path diversity from end to end along with link utilization; ONHS alleviates congestion by at least about 20% across various scenarios.

The mathematical formulation of the problem lends itself to the rigor and sound reasoning of minimum tunnel overlap and path diversity. Its implementation in the VXLAN/IPv6 environment makes it realizable in modern networks.

The paper presents an interesting and technically sound proposal to improve resource utilization in WANs based upon overlay approach with hop-by-hop routing. From simulation, the proposed architecture seems to exhibit better characteristics with respect to path diversity and alleviation of congestion pressure. However, further functional analyses, application-wise performance metrics, and actual scalability evaluation would instill greater confidence, thus facilitating the adoption of the proposed solution in production environments.

 

Recommended improvements

 

1.The paper lacks metrics, or definite measures, to determine computational cost or overhead induced by genetic periodic monitoring of tunnel load, specifically in ONHS. An extended discussion would be in order to increase solution complexity when going toward very large networks.

 

2 While congestion is well handled, the system's behavior is not explored considering sudden failure in overlay nodes. As resilience is a key issue in plight of WAN.

3.Tunnel load is assumed to be a single value, not differentiating for traffic types (latency-sensitive vs. delay-tolerant) or QoS. This simply restricts going directly into environments with mixed requirements.

4. The kind of traffic whose features are being detailed are not described in full detail (whether random, bursty, or based on real traces).
5. The evaluation focuses only on the topology metrics (path diversity, link utilization) but not on application performance (transfer time, jitter, packet loss).
6. It is ONHS that depends upon the periodic gathering of congestion information. Such dependence implies a controller, either central or distributed; however, its role is not sufficiently characterized.

7.Flow lets may get disorderly when sent through multiple paths, and hence encapsulation is proposed as a solution. However, no experimental data is presented to show whether this solution works or how it affects protocols such as TCP.

Issues Not Covered by the Given Solution but maybe covered:

1.The solution concentrates on point-to-point transmissions (unicast). In corporate-wide area network (WAN) circumstances, multicast traffic is relevant and would require an extension of the model.

2.Implementable over IPv6 and VXLAN, yet the proposal does little to consider integration with existing routing protocols such as OSPF or BGP.

3.Considering that the number of overlay nodes influences the computing resources, an analysis of the trade-off between performance and energy consumption would have been interesting.

4.The term "simulated tunnel congestion" was used to describe traffic generation, with no information regarding traffic pattern (CTC vs. bursty), statistical distribution, or realistic workload (e.g., trace-based). Recommendation: Use real traffic datasets (e.g., MAWI, CAIDA) or controlled synthetic workloads.

 

5.Metrics considered the diversity and a valid number of paths. How about some gains in: End-to-end latency, Retransmission rate, Jitter, or packet loss? Recommendation: Integrate QoS/QoE metrics and KPIs typical of WAN networks.

6.While a 500-node graph is tested, no running time or memory usage statistics are supplied.

7.While it deals with congestion, it does not consider failures of overlay nodes or failures of physical links.

Author Response

Thank you for your comments. The author's reply is in the pdf file.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

Authors propose a hop-by-hop multipath routing method for overlay networks, leveraging the routing decision-making capabilities of intermediate nodes. There are some comments as follows.

1. Authors need to provide the full names of MOP and ONHS in the abstract.

2. The format of acronyms should be checked and revised. For example, “LFI (Loop Free Invariant) [5] or DAGs(directed acyclic graphs) [6]” should be revised as “loop free invariant (LFI) [5] or directed acyclic graphs (DAGs) [6]”. “multipath overlay node placement algorithm(MOP) and an overlay next-hop selection algorithm(ONHS)” should be revised as “multipath overlay node placement (MOP) algorithm and an overlay next-hop selection (ONHS) algorithm”.

3. As the background of this paper focuses on the routing and overlay, recent high quality related works should be introduced, such as Hybrid overlay-underlay cognitive radio networks with energy harvesting, IEEE TCOM, which also focuses on overlay and underlay. Authors are free to cite the suggested reference.

4. The novelty and strengths of the proposed MOP and ONHS algorithms should be provided.

5. The difference between the proposed routing algorithms and existing ones needs to be clearly summarized.

6. In related works, authors need to introduced the contributions of references in according with the sequence numbers of references.

7. What are the basis or source for the two key principles?

8. The relationship between overlay layer and underlay layer in figure 1 should be specified.

9. How to determine the values of costs and loads for nodes such as Table 1?

10. The proposed routing algorithm should be evaluated in terms of vital network performance such as delay and throughput.

11. The descripts of simulation results should be more clear and concrete, such as “Figures 11, 12, 13 and 14 show the results of the experiment.”

12. Some references such as [2]-[11] are outdated to some extent, and authors need to update recent high quality works.

Author Response

Thank you for your comments. The author's reply is in the pdf file.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

The authors explain and add new description and new figures based on my comments. 

Reviewer 2 Report

Comments and Suggestions for Authors

Authors have well revised the manuscript.