Date: Tue, 26 Oct 2004 17:53:22 -0400
Message-Id: <200410262153.i9QLrM812149@pennstation.cs.columbia.edu>
From: IEEE INFOCOM 2005 <knightly@ece.rice.edu>
To: wonsuck@research.bell-labs.com
Subject: Your IEEE INFOCOM 2005 paper 1568939590
Cc: "Marina Thottan" <marinat@lucent.com>
Cc: "Ramesh Viswanathan" <rv@research.bell-labs.com>
Cc: "Aniruddha Gokhale" <gokhale@isis.vanderbilt.edu>
Cc: "Amogh Kavimandan" <amoghk@isis.vanderbilt.edu>

Dear Dr. Wonsuck Lee:

Thank you for submitting your paper #1568939590 ('Network Simulation via
Hybrid System Modeling: A Time-Stepped Approach') to IEEE INFOCOM 2005.
We regret that your paper could not be accepted for inclusion in the
INFOCOM program.

This year, INFOCOM received 1419 submissions that underwent a thorough
review process including an on-line discussion for all papers and
discussion of 400 papers at the TPC meeting in Houston. Finally, 244
were accepted for publication. While no review process is perfect, we
tried our best to be fair and diligent.

Attached below and also available at
http://www.edas.info/PaperShow.cgi?m=1568939590 you will find the
feedback from the reviewers, which we hope you will find useful in your
research.  Again, thank you for submitting your paper to INFOCOM and we
hope that you will be able to attend the conference.

Regards, Ed Knightly and Kia Makki (TPC co-chairs)


===== Review =====


*** Recommendation (Your overall rating.): 5:Strong Accept 4:Weak Accept
3:Borderline 2:Weak Reject 1:Strong Reject

Evaluation=Weak Reject (2)

*** Contributions (What are the major issues addressed in the paper? Do
you consider them important? Comment on the degree of novelty,
creativity, impact, and technical depth in the paper.): The paper
discusses how to simulate TCP/IP based networks efficiently, by
exploiting some fluid models to describe the behavior of the sources and
the network. Indeed, fluid models have been recently proposed as
practical approach to deal with the simulation of realistic large
networks.

The main idea of the paper is to run a discrete time simulation in which
the fluid models are exploited to compute the state of the system. For
this reason, the approach is said to be "hybrid".

The idea is very promising, but it is not new.

*** Strengths (What are the major reasons to accept the paper? [Be
brief.]): - fluid model well discussed - simulation results under
non-trivial topologies - paper well structured

*** Weaknesses (What are the most important reasons NOT to accept the
paper? [Be brief.]): - lack of originality; fluid models have been
already developed to model most of all the TCP/queue dynamics. The
hybrid approach is not new. - poor scalability of the proposed
simulation approach, which is not able to deal with very large number of
sources. In other words, in all the scenarios where the proposed
approach works, also NS can be used. - narrow applicability of the
model, since it refers only to infinite length TCP sources with only
drop tail buffers - missing a critical discussion about the proposed
approach

*** Detailed Comments (Please provide detailed comments that will be
helpful to the TPC for assessing the paper. Also provide feedback to the
authors.): The main advantage of fluid models to study TCP-based
networks is that fluid model MAY scale very well with respect to the
complexity of the network.

During the last Infocoms, many papers have addressed this topic, showing
the feasibility of the simulation of very large networks (with thousands
of nodes, million of sources, and very high bandwidths) thanks to fluid
models. See, for example, [*] and [**].

It is very questionable the motivation behind the authors' effort to
design model for a single fluid source, which mimics the complexity of
the Finite State Machine present in TCP. This approach cannot scale. The
simulation results present in the paper refer to networks with hundreds
of souces and nodes. Indeed, the computational complexity of the
simulation is O(num_sources), as NS: this is very clearly shown in the
pseudo-code at page 7.

In addition to "philosophycal issues" which could not be shared by all
the researches, the main techical point missing in this work is the
discussion about the simulation gains with respects to NS, in terms of
simulation time and accuracy. For example, it is unclear how to set the
time step of the simulation, which is a crucial parameter for the final
complexity and the final accuracy of the simulations.

Finally, in appendix B the discussion about the correction parameter
should be compared with similar results found, for example, in [*].

Ref: [*] "SHRiNK: A method for scalable performance prediction and
efficient network simulation", Pan, R.; Balaji Prabhakar; Psounis, K.;
Wischik, D.; INFOCOM 2003

[**] "Flow level simulation of large IP networks" Baccelli, F.; Hong,
D.; INFOCOM 2003.


===== Review =====


*** Recommendation (Your overall rating.): 5:Strong Accept 4:Weak Accept
3:Borderline 2:Weak Reject 1:Strong Reject

Evaluation=Weak Accept (4)

*** Contributions (What are the major issues addressed in the paper? Do
you consider them important? Comment on the degree of novelty,
creativity, impact, and technical depth in the paper.): Simulation of
large scale network including the dynamic rate adaptation of TCP is an
important issue today. This paper proposes some improvements (in
particular fr/fr phase modeling) in the hybrid simulation approach for
the simulation of TCP traffic. It is based on the previous model
background of hybrid system with finite state machine formalism.

*** Strengths (What are the major reasons to accept the paper? [Be
brief.]): Hybrid approaches seem to be indeed promising and relevant for
large IP system simulation. And this paper offers a technical overview
of such a possibility through the modeling choices they did.

*** Weaknesses (What are the most important reasons NOT to accept the
paper? [Be brief.]): They are still many modeling aspects that have to
be revisited and improved. It lakes of some maturity.

*** Detailed Comments (Please provide detailed comments that will be
helpful to the TPC for assessing the paper. Also provide feedback to the
authors.): - The survey of existing approaches is not complete: in
particular, the papers (INFO'02,INFO'03) based also on the hybrid flow
level model (synchronization - billiard) for large scale simulation
should be mentionned. - p.3: The queueing delay of ACK has no reason to
be ignored (except for simplicity of presentation?) - p.5: equation (14)
and and the exit condition of fr/fr seems not relevant: it implies that
if n packets are lost, fr/fr phase will have a duration of at least n x
RTT ? Since ack(t) is a fluid rate, if the ack rate is not zero, it will
take exactly n x RTT ? - p.5: drop "detection" condition is not defined
- p.6: equation (16): condition "if" "ow" missing - p.6: equation (20):
mixing different unities (bandwidth and queue size!) - p.6: equation
(22): I don't see how this formula can work! taking the ceil of the real
value quantity will lead to an arbitraly high drop rate depending on the
time step! - p.7: it is NOT trivial that smaller time step introduces
small error: in particular because of (22), it might be even wrong! -
p.2, col1, l.20: a fluid model - p.2, col2, l.-15: are difficult - p.3,
col1: shouldn't the sending rate = link-out-rate and the arrival rate =
link-in-rate ? - p.3, col2, l.-6: TPC -> TCP - p.4: Mss: maybe better to
use another notation; can be confused with MSS, Maximum Segment Size
which is almost "reserved" notation for TCP


===== Review =====


*** Recommendation (Your overall rating.): 5:Strong Accept 4:Weak Accept
3:Borderline 2:Weak Reject 1:Strong Reject

Evaluation=Weak Reject (2)

*** Contributions (What are the major issues addressed in the paper? Do
you consider them important? Comment on the degree of novelty,
creativity, impact, and technical depth in the paper.): Paper extends
hybrid systems models developed earlier for fast simulations.

*** Weaknesses (What are the most important reasons NOT to accept the
paper? [Be brief.]): The principal weakness of the paper is lack of
evaluation. The approach is useful because discrete event simulators
(a)d do not scale, and (b) they are slow. This paper neither provides a
scenario that discrete event simulators (ns-2) cannot simulate, neither
do they provide speedups of the hybrid simulator as compared to the
discrete even simulators. The authors should also cite a paper by Liu,
Lo Prest et. al., Sigmetrics 2003.