Cache-Enabled Adaptive Video Streaming: A QoE-Based Evaluation Study
Round 1
Reviewer 1 Report
This study addressed the challenge of acquiring video segments that meet the requirements of both the network and the users' Quality of Experience (QoE). The authors conducted simulations using a network of multiple video servers and a video client. They also implemented a peer-to-many communication framework for adaptive video streaming and a video server caching algorithm based on specific criteria aimed at enhancing either the network status or the user experience. There are some concerns that show be considered by the authors:
1- The novelty of the paper is not clear.
2- The authors have not addressed many related studies. For example, in QoE section is superficial and did not cover the state-of-the-art approaches.
3- The details of implementation such as the configuration of the encoder, the network traces, the type of player, and the ABR algorithm are not clear. Why the authors did not use one of the following tools for their performance evaluation: Mininet, NS-3, and CAdViSE [https://github.com/cd-athena/CAdViSE]?
The paper needs to be revised for its English language.
Author Response
Dear Reviewer,
Please find the responses in the attachment.
Best regards,
Eirini Liotou
Author Response File: 
Author Response.pdf
Reviewer 2 Report
The investigation of DASH and caching (as two independent mechanisms) is interesting.
However, the reviewer would like to point out a set of issues and opportunities to further develop the contents of this paper:
1. Upfront clarification of real-life scenarios for which the study is relevant (e.g. for mobile streaming, as expressed in line 311).
2. The authors have chosen chunk lengths of 5 s, which appears to be quite long and might affect both the dynamics ("nervosity") of DASH as well as the durations of stalls. Why has this value been chosen? Have the authors taken alternatives into account (or tested)?
3. As they are used for comparisons, the averages / mean values shown in figures 3-19 and 23-24 should be accompanied by confidence intervals. This implies running replications of the emulation study with different random impacts (e.g. different videos). One single measurement (point) per parameter combination always brings the impression of episodical results.
4. It is unclear to which extent random network impacts (e.g. outages in mobile networks) have been tanken into account. Such random impacts might worsen the achievable utility and "confuse" DASH even further. Currently, this study seems to assume best-case network performance (but the reviewer might have missed something here, but the statement in line 389 points in this direction). For instance, an (approx.) 1:1 relationship in Figure 12 is hardly achievable under "bumpy" network conditions, which are expected to show similar effects as malfunctioning caching. Probably this needs to be left for future work, but the authors should be very clear about any best-(working network-connectivity-)case assumption.
5. Parts of the results (simplified: SWG in general better than GBS, low-rate streams and bigger buffers allow for higher utilisation) are not really surprising. It would be nice to push more for the unexpected results, and reflect on ways how to handle strange behaviours (e.g., vary the length of the chunks).
Some minor items:
- line 107: at => in
- line 265: variables M and N in italics
- lines 307 and 309: better use indices for 1 and 2
- line 410: The increase looks logarithmic (until levelling off) rather than linear
- line 478: R1 = 0 is probably not meaningful.
- List of references: Please harmonise the citation style (e.g. full or abbreviated first name, last name first or last)
The reviewer has observed few minor items:
- Line 107: in => at
- 307
Author Response
Dear Reviewer,
Please find the responses in the attachment.
Best regards,
Eirini Liotou
Author Response File: Author Response.pdf
Round 2
Reviewer 1 Report
The authors have replied to my comments. There is not any further feedback.
Author Response
We would like to thank the Reviewer for his/her positive feedback on our revision.
Best regards,
Eirini Liotou on behalf of the authors
Reviewer 2 Report
Thanks for the revision and the replies. Some further observations:
- It would have been nice to motivate the choice of 5 s-intervals (compromise) even in the text :)
- The reviewer is still not convinced regarding the missing confidence intervals. When averages are to be compared, they should be standard... If several measurements were averaged and the intermediate results were still available, it would be possible to calculate the standard deviation or the standard error (and derive confidence intervals from the latter), to give the reader some indication of the eventual variability of the results and the significance of any difference. Alternatively, a t-test could be performed...
By the way, lack of time should is not really an argument... things have to be sound ;)
- Figures 15-17: R_1 is used on both axes (for different measures), which is confusing. Please consider using a different notion for the measured average rates (e.g. M_1 and M_2)? (_ indicates subscript). BTW, Excel allows for both italics and subscripts even in the axis names and legends. It is nice to use the same kind of notation all over the place.
- Typos in header texts of Figures 18+19: "Trhoughput" => "Throughput"
- Please harmonise the use of a blank between number and unit (e.g. L = 100 MB).
- There are a couple of broken references (p 3, p 6 and twice on p 10). Please correct.
Author Response
Dear Reviewer,
We would like to thank you for the second revision of our paper. Please find our responses in the attachment. Looking forward to your feedback.
Best regards,
Eirini Liotou
Author Response File: Author Response.pdf
Round 3
Reviewer 2 Report
Thanks for the recent improvements.
