Subject: Re: Request to review a paper From: Michael Luck Date: Tue, 27 Mar 2001 10:28:33 +0100 (BST) To: Geoffrey Fox Geoffrey, I append my review for C496: Open Consensus by Romain Boichat, Svend Frĝlund, and Rachid Guerraoui below. Best wishes, Michael Luck. -- Dr Michael Luck Electronics and Computer Science voice: +44 23 8059 6657 University of Southampton fax: +44 23 8059 3313 Highfield email: mml@ecs.soton.ac.uk Southampton SO17 1BJ http://www.iam.ecs.soton.ac.uk United Kingdom C496: Open Consensus by Romain Boichat, Svend Frĝlund, and Rachid Guerraoui > D: Referee Comments (For Editor Only) It's difficult to see this as more than incremental advance that is too closely tied to the previous work. This may not be a problem, but the authors need to do more to ensure that the contribution is clear, etc. Recommendation: reject Open Consensus Boichat, Frolund and Guerraoui The paper offers a "reshaping" of consensus to make it effective in practical crash recovery models. The motivation for the work seems sound, and the contribution valuable, but the authors need to do more to show that this is so. In particular, it isn't easy for the reader to understand the improvement and advance over existing work because the presentation doesn't lend itself well to this by omitting an appropriate level of background coverage. (This is especially important in a paper such as this which offers incremental advance over previous work, and is intimately tied to it.) There are also some other weaknesses as described below. In short, there is value in the paper, but it needs some significant reworking. The paper is based heavily on, or relies is strongly intertwined with, the earlier paper by Aguilera, Chen and Toueg. This is not unreasonable, since both papers deal with very similar problems, but the seeming reliance of this paper on that suggests that it needs a much more extensive review of that work. The emphasis on ACT00a, with numerous citations underlines its importance, yet the exposition of that work is limited. Throughout the paper, reference to, and comparison with, ACT00a is made, yet there is no point at which this is coherently presented. In fact, this is a general point to be made in relation to a body of work --- there is generally insufficient background to provide adequate context and understanding of the key motivating and informing prior and related work. This is particularly evident eg on page3, where the dicussion of reshaping consensus is inevitably tied to existing work that is inadequately introduced. The assumption is made that the reader will necessarily be familiar with all aspects of the cited work (eg Lamport) and that reference to specific aspects of that is acceptable. This is not in general a good assumption to make, and further elaboration is required. The limited code descriptions at the end of the paper didn't seem to offer very much at all, and it wasn't obvious to me why they were included. There are also several minor errors that cloud the clarity of the paper. p3, line 26. Rephrase to avoid the double negative. p4, line 18. looses -> loses p4, line 24. looses -> loses p4, line 25. what are the channel definitions of ACT00a? p7, line 2. well-behaviour -> good behaviour? p7, first line of figure: should this be v1' or v'1? It seems to conflict with the text. p7, line 13: "a process decides another value that it proposed even if this value was not decided" seems not to make sense to me. p7, line 15: even this -> even though this p8, line 8: should "proposee" be "proposer"? p8, footnote: rotatcing -> rotating