REVIEW: "King: Estimating Latency between Arbitrary Internet End Hosts"

REVIEW: "King: Estimating Latency between Arbitrary Internet End Hosts"

The paper proposes a technique for measuring internet latency between
abitrary hosts, without the need for explcit cooperation from neither the
end hosts nor infrastructure in the internet. It uses DNS servers as non
cooperating infrastructure. King works by resolving the authoritative
nameservers of each end host A nd B, measuring the latency to a name
server, say NS_A, and later request a recursive query at NS_A of the
opposite end host B. This leads the server NS_A to query at the opposing
name server(NS_B) and therefore by subtracting the initial round trip time
taken to NS_A from the time taken for the query result can be taken as the
RTT time between the dns servers, and by extension the RTT between the
arbitrary hosts A and B. The assumption that King makes is that hosts are
located near their DNS servers.

A major stregth of King is that it requires no landmarks to be deployed in
the internet nor it requires any explicit cooperation from the end hosts.
The added ability to "trick" name servers to forward qeries to desired
server improves the accuracy of King as it gives guarantees that the
latency being measured is that between the desird nameservers. By using a
service already widely available without disrupting its functionality
gives a it means that DNS maintainers will have no incentive to disable
access by King, as long as their overhead is not significant. As shown in
figure 11, paths to end hosts and those to their name servers often
overlap considerably which yelds high accuracy.

At best, King is expected to wok 94 percent of the time. Although this
accuracy is good for most cases, it may not be appropriate for measurement
studies in which accuracy is prime. Whenever hosts are not near name
servers, the accuracy of King deteriorates. It is not uncommon for large
ISPs to provide centralized DNSes such as Sympatico DNSes servicing their
customers in provinces of Ontario and Quebec. Although it was found that
79% of hosts have at least one recursive authoritative nameserver, it is
not taken into account that such nameservers are at times not located near
the end hosts. For example, many universities establish secondary
nameservers at non local institutions as for redundancy purposes.
Received on Thu Nov 03 2005 - 10:04:34 EST