Amit Saha:

Title: Combs, Needles, Haystacks: Balancing Push and Pull for
Discovery in Large-Scale Sensor Networks
Authors: Liu, Huang, Zhang
SenSys 2004
----------------------------------------------------------
                                                                                
Abstract:
-------------
The authors propose a compromise between totally pull and totally push
strategies and propose a middleground. In the proposed approach the
sensors which experience an event send this information "vertically"
to some of its neighbors (both down and up) and the query is sent (say
by the root node) and the query proceeds in "horizontal" stripes and
hence the query just has to meet some node in the vertical neighbor
list of nodes which had originally experienced the event. Location
information is assumed at all sensor nodes.
                                                                                
Comments:
----------------
The paper as such presents a neat idea. However, I find the following issues
                                                                                
1. The proposed mechanism cannot handle "urgent events" which cannot
wait for the next query to come by. Why cannot these "urgent events"
travel the tooth of the combs to reach the root. The root for such
urgent events can be fixed. Travelling the tooth of the needles is
still better than routing towards the root directly on every event.
                                                                                
2. The evaluation is very weak an sloppy. Is 'W' and 'w' the same parameter ?
                                                                                
3. The authors introduce Figure 10 and then add the sentence  "This
result  matches well with the predication in Eq. (8)" Given that the
word is "prediction" and not "predication" this is a trick on the part
of the authors since the sentence does not mention Figure 10 and
Figure 10 does not have any notion of "analytical results" plotted in
it.
                                                                                
4. What is the unit of Y-axis on Figure 10, is it normalized ?
                                                                                
5. What is the unit of 'w' (0.5, 0.6, 0.8), etc. The readers are asked
to "note that when 'w' goes to 1.5, the combing is similar to
flooding". Neither Figure 12, nor Figure 13 (for that matter none of
the figures) actually have any line for w=1.5.
                                                                                
6. Figure 14 is a trivial graph in wireless network. Place a set of
node randomly in a square area and use some probability 'p' of node
failure, and voila, you have Figure 14. It is just a place filler, can
be used to motivate something, not to evaluate. Besides what was the
'w' value for Figure 14.

To reiterate, the paper idea is neat but the evaluation is terrible
and could have been done in a much better way.


*********************************************************************

Andreas Haeberlen:

The authors describe a new family of query techniques for sensor networks.
These techniques are based on rendezvous between queries and data, i.e.
the data is replicated on a subset of the nodes and the query on another
subset; the replication is such that the two subsets are small but
intersect in most cases.
                                                                                
The basic idea behind this paper is not particularly novel; a simple
variant was described in 1999 by Stojmenovic and Vukojevic (University
of Ottawa SITE TR-99-09), but this is not mentioned in this paper.
Credit should be given.
                                                                                
Apart from that, the technique is based on assumptions about the sensor
network that are somewhat simplistic. For instance, I doubt that the
technique would work well in a real deployment with obstacles, sensor
failures and an irregular deployment pattern - even if redundancy is
increased, as described in Section 4. The dilemma here is that to increase
reliability, more energy has to be spent. Another question is whether the
communication pattern imposed by this technique would conflict with
other energy-saving techniques such as sleeping; I think this matter should
be discussed in more detail.
                                                                                
Finally, it is not clear to me whether it even makes sense to trade off
reliability for energy in the scenarios listed in the paper. If there is
a tank on the battlefield, a failure probability of 15% for queries is
just not acceptable. I think the paper would benefit from a more realistic
application scenario.


****************************************************************************