Sting Paper Review

 The paper is introducing Sting, a measurement tool which can be deployed on
a host to measure the packet loss in the forward and reverse directions. The
main contribution of Sting, in comparison to its predecessors, is its
accuracy on measuring the packet loss rate and also in that it doesn't need
its user-level software to be deployed in both sides, namely the receiver
and the sender.
 KEY STRENGTH(s) :
 The key strength of the paper, to my view, is its simple exploitation of
TCP features to attack its restrictions. More specifically, author's use of
the retransmit feature to provide fast ack arity, and also the property of
TCP that ignores buffering overlapped bytes which enables the author to
measure packet loss rate for data bursts among other ideas are significant,
since these ideas help to build a system which utilizes TCP (which is widely
deployed) instead of some un-deployed fact in the Internet.
 The paper's claims are nearly well-tailored to its results, so it seems
hard to criticize it at least for the necessity of its results against its
claims. The only thing I can mention is that the authors are not explicitly
showing evidences for empirical correctness of their tool, rather they are
saying that our method worked for a simulated network. I think it is
necessary to provide a way of proving this tool's correctness in the real
Web settings, and it is actually possible since, as the authors mention,
there are off-the-shelf measurement tools which are deployed at both the
sender and receiver hosts and can measure the packet loss rate exactly. So,
this experimental evaluation can be done and is more justified than their
simulations.
 As a future work, I think this paper can suggest means for measuring
properties of Web-Servers, since it is the first method which allows passive
measurement of such in-directly participating hosts in the Internet. As a
suggestion, it seems to be possible to measure the traffic of these web
servers as well, using this tool. This idea is based on the fact that as the
traffic of a web-server goes higher, the packet loss rate also increases
(especially in the reverse direction) and from the ratio of these two
values, may be one can deduce the number of hosts or the bandwidth of a
web-server like Google!
Received on Mon Oct 31 2005 - 11:06:24 EST