It is a paper of good coverage. But every part of it seems mediocre.

Review: TCP Nice

Reviewer: Di Niu

It is a paper of good coverage. But every part of it seems mediocre.

Inspired by the popularity of background transfers, Nice is proposed
to reduce interference with demand flows and reap available spare
network bandwidth. Nice is primarily a modification to Vegas.
Theoretical analysis, ns-based tests and application-specific case
studies are provided to show the advantage of Nice over Reno or
Vegas. The paper represents a complete work and covers a good range
of topics. However, in solving a pretty easy problem, any part of it
seems to be just mediocre.

First, it is not concentrated enough and contains much vagueness.
Nice is inspired the problem that background transfers could
interfere with foreground transfers. And the fundamental goal of Nice
is to eliminate such an interference. However, it is not clear what
background transfers really are or why interference incurred by
background transfers is an important problem. In Section 6, the paper
studied about HTTP prefetching. However, it is not clear enough why
prefetching is related to the backgournd transfers or why prefetching
is important to study. In this sense, the paper is not focused enough.

Second, the theoretic analysis contains too much irrealistic
assumptions, and thus it is not very valuable. For example, it
assumes that "at the end of each RTTepoch, a nice sender accurately
estimates the queue length during the previous epoch." This
assumption alone could kill the analysis, not to mention the
synchronized flows and infinitely small packets. Actually, theorem 1
should not be called a theorem at all, as the analysis is largely a
contrived effort to come with some good properties of Nice. A small
revision of the assumptions could lead to a totally different result.

In fact, Nice is primarily a modification to Vegas and thus its value
is quite limited. In order to support a very simple idea, that is to
reduce flow interference, the paper struggles to have a good
coverage, trying to put a lot of loosely correlated things together.
Unfortunately, every part of it just turns out to be a mediocre stroke.
Received on Thu Oct 05 2006 - 10:37:29 EDT