CSC2231 Review of King

By taking the advantage of DNS architecture, King can estimate the latency
between arbitrary Internet end hosts. This is useful since ping is often
forbidden by many network administrators.

However, their dependence on DNS also limits its accuracy and confidence
in some degree. We could consider it as a good approximation of end-to-end
latency, but we never know how close it is to the real latency. The
fundamental reasons are as follows: First, they assume that a host is
close to its ADNS. It may be almost true, but we do not know when it is
not the case. Second, DNS query time is also affected by name server
workload. So a heavy loaded DNS can return a long latency. But King will
always think it as the network latency. Third, there may be different
routing paths between end-to-end hosts and the corresponding name servers.
Although the experimental results show quite good accuracy, we are not
sure when the inaccurate estimation will happen due to above factors.

In addition, since lots of Internet hosts are behind NAT or proxy, and
thus they do not have a valid domain name. In this way, King can only
estimate the latency between the request host and the NAT of the target
host. This may increase the inaccuracy of King since the host could be far
from its NAT server or proxy.

In brief, King is a good tool to roughly estimate the end-to-end latency.
But it is really not an end-to-end host measurement. Maybe we have to
install some specific software between two hosts for accurately measuring
the end-to-end host latency.
Received on Thu Nov 03 2005 - 10:42:51 EST