ViewVC logotype

Contents of /bcast-kanada/reviewer-comments

Parent Directory Parent Directory | Revision Log Revision Log | View Revision Graph Revision Graph

Revision 1.2 - (show annotations)
Wed Mar 29 09:18:27 2006 UTC (12 years, 10 months ago) by riso
Branch: MAIN
CVS Tags: sirocco06-cameraready, HEAD
Changes since 1.1: +17 -8 lines
Corrected typos reported by the 3rd reviewer (Richard)

1 1:
3 The authors introduce a new model for dynamic link failures in the
4 synchronous setting. It encompasses two previously studied deterministic
5 models that the authors argue are problematic for opposing reasons.
6 The first model limits the number of messages lost to an absolute bound
7 per time step. This encourages sending a large number of messages. The
8 second model allows a fraction of the messages per step to be lost. In
9 some situations this encourages sending a small number of messages (since
10 one is guaranteed to be delivered). Their model combines the two by
11 using two parameters, one a threshold the second a fraction, where the
12 number lost is max of threshold and fraction of message sent. This has
13 the potential of avoiding the two extremes above and of concentrating on
14 algorithms in between.
16 Having defined the model, the authors give a large number of results
17 concerning broadcasting in this model on particular networks as well
18 as more general results. Some of these results may involve arguments
19 familiar
20 from the previous settings but some appear to use new techniques. In
21 either case, the arguments are non-trivial.
23 The organization and writing are fairly clear. The new model is likely
24 to generate a lot of further work as the authors leave plenty of
25 interesting
26 open problems.
29 ===========================================================================
31 2:
33 - The paper only considers the special case that alpha = 1 - epsillon and
34 T=c(G). The extensive discussion of the more general case (any T) in the
35 abstract, and the more general model (any T, any alpha) in the introduction
36 and model is only misleading and confusing. I was confused by this,
37 believing that you would use the same T to compare different topologies,
38 and hence couldn't understand why the ring algorithm cannot be used in a
39 complete graph by sending only on the links of an embedded ring.
41 The paper would be much better off focusing a priori on the single model
42 considered therein without the lengthy (misleading and confusing)
43 discussion of the more general models. The paper is already dense as is,
44 and hence some effort should be dedicated to helping the reader follow it.
46 If you want to define the more general model as a direction for future
47 work, I suggest that you do so at the end of the paper in a
48 "discussion/future directions" section.
50 - Since your draconic model does not remedy any of the problems of previous
51 models so extensively elaborated in the introduction, this whole
52 discussion should be removed.
54 Instead, focus the introduction on the model you do study, the challenges
55 in this model, and the key ideas of the results.
57 - In the abstract, and in the model paragraph, you say that more than T
58 must be sent in order to ensure that one is received. Later it turns out
59 that if you send T this ensures that one is received.
61 - It is important to explain how the considered simple threshold model
62 differs from the previously defined cumulative threshold model.
64 - The introduction makes some claims about consensus, which are of no
65 relevance to the current paper (dealing with broadcast). The focus on
66 synchronous models is motivated by the inability to solve consensus in
67 asynchronous ones. However, reliable broadcast *is* solvable in
68 asynchronous systems. The introduction further claims that broadcast has
69 "immediate consequences for ... k-agreement.. ".
70 What immediate consequences? AFAIK, k-agreement is unsolvable in some
71 systems where reliable (albeit not atomic) broadcast is solvable.
73 Again, the paper would greatly benefit from focusing the introduction on
74 the context of the current paper without making confusing (and imprecise)
75 statements about other models and problems.
78 ===============================================================================
80 3:
82 Add the word "fault-tolerant" or something similar in the related theorems.
84 page 2 lines 3 and 4 the word "drawback" has been repeated twice
85 line -6 better explain why exactly n-1
86 page 6 it is not clear why Algorithm 1 should be working. Please explain it
87 in better details.
88 line 5 "send": who sends these messages?
89 line 12 assume n=5, cycle of line 2 goes from 1 to 2.
90 consider k=2, thus d=2
91 we are then in the else condition and
92 x=4-4=0, y=2-0=2
93 at line 12 you send messages to vertices n-x=5-0=5
94 but this vertex does not exist. Please fix
95 I looked at the proof in the appendix and there you assume
96 that x>=1 (page IV line -3) which here does not hold
97 unless I made a mistake.
99 page 7 better explain the meaning of the different invariants.
100 E.g., why I1 implies that the last delivered token is green?
101 Do you mean sent on {a,b}. Could it not be red? A reader should be
102 convinced of its truth without reading the appendix.
104 Explain more the general idea of algorithm 2.
105 line 6 is B=empty at the very beginning?
107 line -5 how do you select P?
108 page 10 what is Pz? Where does z appear?
109 Again better explain algorithm 3.
111 =========================================================================
113 Done:
115 1:
117 2:
119 3:
121 line 19 proposed -> propose
122 line -6 is a -> in a
123 page 4 line -13 in the -> at the
124 page 8 algorithm 2 line 3 has -> have
125 page 9 section 4 line 1 we assume -> we consider k-connected graphs
126 and we assume
127 line 7 a dot is missing
128 page 12 please extend the name of the journals in [6], [11]
129 [13] complexity -> Complexity

CVS Admin">CVS Admin
ViewVC Help
Powered by ViewVC 1.1.26