Review - Sting - a TCP-based Network Measurement Tool

From: Ian Sin <ian.sinkwokwong_REMOVE_THIS_FROM_EMAIL_FIRST_at_utoronto.ca>
Date: Sun, 30 Oct 2005 13:02:05 -0500

This paper introduces Sting, a packet loss measurement tool that uses
TCP. Sting can get accurate information and discriminate between forward
and reverse direction loss rates in TCP traffic and thus makes it
superior to its ICMP-based counterparts.

This work is important because web applications react differently to
loss in forward and reverse packet losses and there are no tools that
have successfully been able to achieve that. The proposed ideas to use a
TCP-based measurement tool are insightful as it uses existing TCP
behavior to derive useful information. The widely deployed TCP stack
does not have to be modified and re-deployed to support this measurement.

The weakness of this paper, in my opinion, lies in the presentation of
its evaluation section. I am not sure that Figures 6 and 7 give a lot of
information and the fact that they picked one web server arbitrarily is
not very convincing. Do other web servers have the same distribution? Is
the one they picked a popular one or a random one? They seem to conclude
from these graphs that ALL web servers have reverse loss rate twice that
of the forward loss rate. I believe that they could have done a better
job with evaluating their tool.
Received on Sun Oct 30 2005 - 13:01:06 EST

This archive was generated by hypermail 2.2.0 : Sun Oct 30 2005 - 20:50:54 EST