Reviewer: Michela Becchi
Date: 10-6-2005
How would you rate this paper, relative to others we have read? bottom 50%
How would you rate your knowledge of the topic of this paper? novice
What problem or issue does the paper address? Why is it important?
The authors of the paper propose an abstract model called Single-Buffered router and use it to classify different router architectures and in particular to analyze the Parallel and Distributed Shared Memory routers. In particular, one objective is to find the conditions under which these two architectures approximate the behavior of an ideal (FCFS and PIFO) shared memory router. At the end they compare a DSM router with a CIOQ router as well.
What are the main contributions of the paper and why are they important?
I must say that I am not familiar with this area. Anyway, I think that the main contribution consists in the application of the Constraints Sets methodology to the study of the Parallel and Distributed Shared Memory routers. This leads to the theorems listed in the paper itself, which give the equivalence of Parallel and Distribued Share Memory routers and Crossbar-based DSM routers to FCFS and PIFO shared memory routers in terms of memory bandwidth and speed.
How significant are these contributions relative to previous work?
I guess that this paper gives a contribution in that the presented analysis had not been previously performed. Anyway, as discussed in the "Practical Considerations" section, I am not sure about the practicability of DSM routers. In particular, the following open issues seem to be not negligible: 1)sizing of the buffers; 2) complexity of the algorithm to decide the memory where the arriving packet must be written into; 3) fact that such algorithm must run centrally to have global knowledge about the packets.
Give detailed comments justifying your view of the paper.
The paper applies Constraint Sets analysis to Parallel and Distributed Shared Memory routers in order to determine under which conditions (in terms of memory bandwidth and speed) they can emulate ideal shared memory routers. The analysis is theoretical. At the end the authors discuss the feasibility of such systems and do some comparisons with CIOQ routers.
I must say that I found quite difficult to read and follow this paper. I am certainly novice to this topic, but anyway I did not like at all the way it is presented. In my opinion the authors do not clearly state in advance the goals of their analysis, and they even lack to summarize their findings in a conclusion section. In many points I had the feeling that the paper provides with a collection of not-so-structured pieces of information.
Since the paper provides a great amount of theoretical results, I think it would have helped if the central section (section II) had first anticipated the goal and the results of the study, and then had gone into details with the analytical part. Somehow it seems that even the explanation of the Constraint Sets intuition and methodology comes out from nothing.
I found also very strange the way this paper ends: 1) it does not have a real conclusion section; 2) it references work only about the application of the Constraint Sets analysis; 3) The "Open Problems" point the the VI section seems to be an isolated point: should it represent a future work part? [2] lets think that the real contribution of the paper itself is just the application of this methodology to shared memory routers.
Finally, I don't feel I have enough knowledge to judge the analytical part, which seems to be the real contribution of this work. I appreciated the fact that the authors tried to evaluate the feasibility of the DSM routers in section VI, and to enrich the analysis with practical considerations.
I would therefore rank the paper in the bottom for the presentation and organization, even if the correctness of the analytical part would bring it to middle positions.