======= Review 1 =======

> *** Recommendation: Your overall rating (Please try giving as few borderlines as possible).
Likely Reject (top 50% but not in top 30%, needs more work) (2)

> *** Strengths: What are the major reasons to accept the paper? [Be brief.]

The authors have implemented a tool for monitoring congested
network links.

> *** Weaknesses: What are the major reasons NOT to accept the paper? [Be brief.]

The paper is not relevant to ICNP since it is about monitoring
networks rather than protocols. There is no comparison against
the large body of related work.

> *** Detailed Comments: Please provide detailed comments that will help the TPC assess the paper and help provide feedback to the authors.

Is ICNP the right venue for this paper? There doesn't seem to be much
about protocols. It seems this would be better suited to IMC.

How many runs did you do for your EmuLab experiments - I don't see any
error bars.

As you note in your related work section, there is a huge amount of
work in this area. So did you try to compare your tool against any of
the other schemes?

All of the figures should be bigger - they are quite hard to read.

The text in Fig 1 and Table I is jumbled.

There are a few typos, e.g. "PlantLab"

======= Review 2 =======

> *** Recommendation: Your overall rating (Please try giving as few borderlines as possible).
Accept if room (top 30% but not top 15%, borderline for ICNP) (3)

> *** Strengths: What are the major reasons to accept the paper? [Be brief.]

The paper present an interesting approach for queuing delay inference in a multi-hop path. The authors explore several heuristics and have done a measurement survey to validate their approach.

> *** Weaknesses: What are the major reasons NOT to accept the paper? [Be brief.]

The paper present to many heuristics without cleanly validating them. Unfortunately the authors fail in presenting a convincing validation of their works.

> *** Detailed Comments: Please provide detailed comments that will help the TPC assess the paper and help provide feedback to the authors.

As a main criticism of paper is that it develop to much concepts without ensuring that each one is correctly motivated and validated.

The paper begin first by defining three possible situations for coordinate probing and define a fourth one containing all other situations. It states that when we are in the fourth situation good accuracy for the proposed method cannot be obtained. After that much works is devoted in developing heuristics for dealing with the three cases of interest. However it is not before the validation that one figures out that in fact 36% (more than a third) of links that falls in the fourth class, where the inference is not tight ! This percentage make the usage of the approach in operational environment almost impossible as the likelihood that a link of interest falls in this class is very high.

It would have been interesting to just focus the paper on inferring the queueing delay with the proposed heuristics without going into the monitoring that is unfortunately explained in a wooly way.

For example no explanation in the paper is given for the linear combination argument used for the 4 pair case in table 1, where it seems to be an important case.

The congestion intensity used in Fig. 3 is not defined !

The use of logarithm binning for the delay distribution is not motivated. Taking into account that the binning have an important impact on the threshold obtained thereafter, the authors are not giving enough arguments for using this binning.

The authors use frequently the term "typical". It is well known that finding typical pattern in network is very rare. Using this term should be much motivated.

======= Review 3 =======

> *** Recommendation: Your overall rating (Please try giving as few borderlines as possible).
Likely accept (top 15% but not top 5%, significant contribution) (4)

> *** Strengths: What are the major reasons to accept the paper? [Be brief.]

The idea proposed for identifying the locations of congested links
is fresh and interesting. The extensive experiments in Emulab and
over the Internet reveal maturity of the work and effectiveness of
the advocated method.

> *** Weaknesses: What are the major reasons NOT to accept the paper? [Be brief.]

The paper does not show experimentally that the proposed method
is better than existing techniques such as Tulip and Pathchar.
The Introduction gives a "marketing" feel by not focusing on specific
technical features and contributions.

> *** Detailed Comments: Please provide detailed comments that will help the TPC assess the paper and help provide feedback to the authors.

The submission proposes and evaluates Pong, a tool for identifying
the location of congested links. Using queuing delay as a congestion
signal, Pong locates congestion spots with an innovative technique
of sending probes from both ends of a tested path. Pong also
generalizes this interesting technique to work in more complicated
scenarios.

The thorough evaluation includes experiments in Emulab and PlanetLab.
The work clearly shows its maturity and many months of efforts.
The attention to the measurement accuracy is highly commendable.
The presented results confirm the effectiveness and low overhead of
Pong.

The related work positions Pong clearly among the prior proposals.
Tulip and Pathchar seem to be the closest existing tools. The
paper will greatly benefit from reporting extra experiments
that substantiate the claim that Pong achieves a higher
accuracy with a much lower overhead than the provided by Tulip and
Pathchar.

The presentation is excellent overall. The paper has a sound
structure and is written lucidly. One gripe is that the Introduction
has a "marketing" feel; it would be better to shift the emphasis
from "selling" the proposal onto specific technical features
and contributions of Pong.

I recommend acceptance of the submission.

======= Review 4 =======

> *** Recommendation: Your overall rating (Please try giving as few borderlines as possible).
Accept if room (top 30% but not top 15%, borderline for ICNP) (3)

> *** Strengths: What are the major reasons to accept the paper? [Be brief.]

The paper presents a novel (and interesting) method for assessing the degree to which specific links in the network are congested. Results of emulab- and planetlab-based studies are presented.

> *** Weaknesses: What are the major reasons NOT to accept the paper? [Be brief.]

There is no comparison to any of the many existing methods. The authors' own metric suggests that the measurements were not reliable on over 1/3 of the links.

> *** Detailed Comments: Please provide detailed comments that will help the TPC assess the paper and help provide feedback to the authors.

This paper contains some very nice ideas: the use of "path patterns" to determine which measurement scenario applies; the idea of a "measurability score", which gives an idea of the reliability of the results. The inclusion of PlanetLab measurements is also a plus.

However, as always with attempts to solve such measurement problems, the issue of "ground truth" is a challenge. The use of emulab experiments to validate the congestion intensity metric is good. The measurability score, while a great idea, is more problematic since it constructed in a pretty ad hoc fashion. The paper would be greatly improved by including some emulab results to evaluate the "measurability score".

======= TPC Review 5 =======