King: Estimating Latency between Arbitrary Internet End Hosts
-------------------------------------------------------------
K. Gummadi et al.

The authors propose a new tool for estimating network latency between 
two hosts. In measuring this latency they query the nameservers associated 
with the respective host and they end up determining the latency between 
those nameservers which they take as an approximate value of the one 
between the hosts.

The paper strength comes from the idea used in the tool. Instead of 
deploying landmarks and/or requiring hosts to do computation like 
previous approaches - IDMaps and GNP, King requires no host cooperation 
and uses an already deployed infrastructure, the DNS. Because of these 
things anyone can use this tool to determine the network latency 
between any two hosts.

However, King's accuracy depends on some issues. Finding nameservers 
close to the host is one source of inaccuracy. Non-recursive nameservers
can be also a problem, though as the authors mention and as I found out
on my own, a lot of the nameservers out there are recursive. One other 
source of errors is the variation in the last hop (hosts connected 
through modems, DSL, ...). I think that this is the main source for poor 
estimates.

Maybe the measurements would be more accurate if we could easily estimate 
the latency between the King host and the last common router and also 
between the King host and the target hosts and nameservers. In this way we 
would actually be able to better estimate the latency between the hosts.