Sting: a TCP-based Network Measurement Tool

From: Jing Su <jingsu_REMOVE_THIS_FROM_EMAIL_FIRST_at_cs.toronto.edu>
Date: Thu, 27 Oct 2005 15:18:51 -0400

This paper describes a method of using TCP congestion control
behaviour to deduce loss on forward as well as reverse paths. Current
tools such as ping and traceroute use ICMP packets, which are often
selectively filtered or rate limited, thus reducing their usefulness
in measuring real link loss characteristics. The key to this paper's
technique is using the ack mechanism for determining whether the loss
occurred on the sender side or receiver side. This technique gives no
information on route taken or at which link the loss occurs at. It
simply gives good loss rate estimates for forward and reverse paths.

It is unfortunate that this technique cannot be implemented easily as
a user-space tool. Undoubtedly this reduces its adoption and use.
The paper hints at perhaps using firewall rule state-machines for
implementing this technique, though I haven't been able to find such
an implementation online. It is also unfortunate that this technique
cannot pinpoint the link location where losses occur.

On page 7, second column, the paper discusses the asymmetry of loss
ratios for forward and reverse paths. Since it is common for paths to
be asymmetric and the reverse path to contain more traffic (as in the
case of web behaviour), it is reasonable for the reverse path to
experience more loss. However, the rate is much lower than what I
expected and the differences were not as large as I expected. 1.5%
may be more than twice of 0.7%, but it's still only a difference of
0.8%. Less than one percent.
Received on Thu Oct 27 2005 - 15:18:55 EDT

This archive was generated by hypermail 2.2.0 : Fri Oct 28 2005 - 22:06:32 EDT