Jin Jin's Review

Summary of the paper

This paper proposes a congestion control mechanism for background
transfer. The current background transfer approaches have several
flaws, such as complicating applications, being too aggressive and
interfering with other applications, and being too timid and not
gaining the benefits of background transfers. Authors introduce TCP
Nice whose goal is for the operating system to manage network
resources in order to provide a simple abstraction of near zero-cost
background transfers.

Nice draw inspiration from TCP Vegas, and its extension adds three
components to Vegas: First, a more sensitive congestion detector.A
Nice flow monitors round-trip delays, estimates the total queue size
at the bottlenect router, and signals congestion when this total
queue size exceeds a fraction of the estimated maximum queue
capacity. Second, multiplicative reduction in response to increasing
round trip times.If the congestion condition does not trigger, Nice
falls back on Vegas' congestion avoidance rules. If a packet is lost,
Nice falls back on Reno's rules. The final change to congestion
control is to allow the window sizes to multiplicatively decrease
below one. By allowing the window to go below one, Nice retains the
non-interference property even for a large number of flows. Both
analysis and experiments confirm the importance of this feature: this
optimization significantly reduces interference, particularly when
testing against several background flows.

Authors provide the analysis and experiment both for ns controlled
test and Internet microbenchmarks. Through them, paper comes to the
conclusion. This new mechanism could almost eliminate interference
with demand flows and reap significant fractions of available spare
network bandwidth. Nice provides a mechanism to improve application
performance by harnessing network spare capacity in a non-interfering
manner. The case studies authors provide demonstrate that Nice can
simplify application design by eliminating the need to hand-tune
parameters to balance utilization and interference.

Points in favour or against

The paper is well written, with fine and clear presentation.
Obviously, it improves the previous work, especially the Reno and
Vegas. But this paper mainly focuses on background transfer. The
primary goal of this paper is to ensure minimal interference with
regular network traffic; though high utilization is important, it is
a distinctly subordinate goal in the algorithm. Here is a problem
that if this algorithm could improve the performance. Authors did not
analysis which one is more important, high utilization or minimum
interference. How to balance these two goals? I suggest authors could
solve this problem at first theoretically.

Moreover, authors did not analyze the fairness and stability problems
which are very important. They are the third and fourth goals?

Another problem is how to choose or calculate the the threshold 't'.
In the paper, authors did not analyze it very clearly. I think it's
very important, because it decides how sensitive the congestion
detector is, and it could affect the algorithm performance.

A good point I want to mention is that this approach is very simple
and easy to deploy. It's a good algorithm for engineers.
Received on Wed Oct 04 2006 - 19:09:49 EDT