Reviewer: Mike Wilson
Date: 10-6-2005
What is the consensus rating for this paper? bottom 50%
How confident are you that the consensus judgement treats the paper fairly? reasonably confident
How would you rate the overall review process for this paper? thorough
How did the disagreements among the reviewers get resolved? Or if you were unable to reach consensus, what were the main sticking points?
The reviewers generally agreed.
One of the few points of disagreement involved the quality of the writing. Most reviewers found the paper very difficult to read. The major complaint was that the authors did not make a clear, explicit statement of the contribution of the paper, nor was future work discussed in any detail. The other complaint was about forcing the reader to flip back and forth between sections to see the proofs associated with some of the theories. The general consensus on this point was that this is actually a good thing. The details of the proof s distract from the points the authors wish to make, and are best deferred until the end.
The consensus in the audience was that there are major problems with single-stage buffering in practice due to the dual requirements of global knowledge and a central interconnect scheduler. Even granting that these exist, the speedup required to achieve the guarantees is still much higher than for dual-stage buffered routers/switches (6x as opposed to 2x).
The authors neglected to consider how their routers/switches work when the requirements for their guarantees are not met. (What if we only have a 5x speedup? What if the scheduling algorithm only approximates the full centralized algorithm?) Some members of the audience saw this as a major lack, while others merely saw a potential for future work.
One final point of disagreement was on the significance of the paper. The paper does not provide a reason to want to use single-stage buffering; quite the contrary, the results in the paper imply that multi-stage buffering is much more efficient. However, as some members of the audience noted, a negative result is still a valid contribution to the literature.