Recommendations for future
large scale simultaneous DNS data collections
In collaboration with ISC and
OARC,
we held the second large-scale simultaneous data collection event on
January 9-10, 2007. We captured tcpdump traces at multiple
anycast instances of the C, E, F, K, and M root servers.
This collection of the DNS root server traffic was part of a
larger measurement effort, "The Day in the
life of the Internet". CAIDA organized an unprecedented
scientific measurement experiment on the worldwide Internet, in which during
48 hours we coordinated global measurements of Internet traffic and relevant
parameters at many participating locations and links. A
summary describes acquired data
in more detail.
With ISC/OARC permission, we indexed these traces into the
Internet Measurement Data Catalog (DatCat). Researchers will need
to contact the OARC
directly in order to access the data.
We are in the process of analyzing the data obtained from the top of the
DNS hierarchy. We presented
preliminary analysis results at the 8th
CAIDA-WIDE workshop (July 2007, Chicago). The paper is in preparation.
A report comparing 2006 and 2007 observations
of root server traffic is available.
We studied DNS traffic collected over a two-day period on January 10-11,
2006 at anycast instances of the C, F and K root nameservers. Our goal was to
investigate how anycast DNS service affects the worldwide population of
Internet users.
First, for each root instance, we examined client locations and the geographic
distances between a server and its clients to determine whether clients
actually use the instance closest to them. We found that frequently the choice,
which is entirely determined by BGP routing, is not the geographically closest
one. Next, we considered specific AS paths and investigated some cases where
local instances had a higher than usual proportion of non-local clients.
Finally, we found that instance selection by BGP is highly stable: over a
two-day period less than 2% of both C-root and F-root clients and <5% of
K-root clients experienced an instance change. We concluded that overall,
anycast roots significantly localize DNS traffic, thereby improving DNS
service to clients worldwide.
We presented
our results at the IEPG meeting
(July 2006, Montreal)
and published a paper at
PAM 2007.
We indexed the data used for the PAM paper into the Internet Measurement
Data Catalog (DatCat). Interested
researchers should contact the OARC
directly in order to access the data.
In 2002 (prior to anycast deployment), Ken Keys analyzed clients and queries
reaching the A, E, F, I, K and M DNS root servers.
-
Traffic at {e,i,k,m}-root for 26 hours starting 2002-08-14.
-
Traffic at {a,e,f,i,k,m}-root for 7 days starting 2002-08-28.
-
Traffic at {e,i,k,m}-root during the DoS attack of 2002-10-21.