Sting: a TCP-based Network Measurement Tool
-------------------------------------------
Stefan Savage
In this paper the author introduces a TCP-based tool for measuring the
end-to-end packet loss in a network. Sting was proposed to replace common
ICMP-based tools, like ping, because of two problems existing with the
latter ones: loss asymmetry and ICMP filtering.
The strength of the tool is represented by the fact that it only requires
sender cooperation and it accurately reports loss rates in both ways
(sendet-receiver and receiver-sender). Moreover, in spite of the issues in
TCP implementations, sting is based on a straightforward idea.
The weakness of sting lies in its implementation; by imposing a change in
the kernel (even a small one) the tool becomes undeployable globally.
However, if one wants to do measurements or analyze network behaviour, it's
easy to use it. Furthermore, this kind of tool can be integrated in wide-area
services that act depending on network characteristics. Such an wide-area
system has one administrative part which makes it easier to deploy a tool
like sting.
Even though they have the problems stated in the paper, tools that use ICMP
are still widely used (ping, traceroute). I guess it's because of the
implementation that sting hasn't been widely adopted.
Received on Mon Oct 31 2005 - 09:53:47 EST
This archive was generated by hypermail 2.2.0 : Mon Oct 31 2005 - 10:42:10 EST