[AfrICANN-discuss] Measuring Root Server Performance
annerachel at gmail.com
Thu Feb 23 12:45:40 SAST 2012
Measuring Root Server
*By *Daniel Karrenberg* <http://www.circleid.com/members/3167/>
[image: Daniel Karrenberg]
Root name servers are a core service of the Internet (For more information
on root name servers, please see the DNS Root Name Server
As such they receive a huge amount of queries and need to answer reliably
with acceptable delay. The RIPE NCC is responsible for operating one of the
13 DNS root name servers K-root which responds to 10,000 — 15,000 queries
per second. Most root servers are operated as a network of distributed
That means a single IPv4 or IPv6 address is announced simultaneously by a
set of name server "instances" deployed in different geographical
K-root operates 18 instances; You can find a map
http://k.root-servers.org/on the RIPE NCC's website.
VisualK <http://k.root-servers.org/visualk/> is a new tool that monitors
the load of the K-root name server supported by each instance. It further
shows load migrations between pairs of instances over time. ViskalK is one
of the tools our operations staff use to monitor the health of K-root.
The image below is a screenshot of the output of the tool. Each instance of
K-root is represented by two concentric circles:
- The first one, filled with colour, has a size proportional to the
number of queries per second received on that instance;
- The other one, indicated by a dotted line, shows the average load over
the previous 30 minutes. This is used as a reference value.
In most cases, these circles overlap. But in some cases you can see that
the dotted line is much larger, for example at the root name server
instance in Poznan, Poland. This means that something has changed recently:
the number of queries has dropped significantly.
In the image you can also see that pairs of instances are connected by
links (or "tentacles") if they are considered topologically adjacent. Links
between root name server instances are generally invisible, but become
active when traffic migration is detected: colour and size of the link
indicate the origin and volume of traffic flow, together with bubbles
pouring into the instance receiving the traffic. In our example you can see
that some load has moved from the instance at NAP (in Miami, Florida) to
the one located at LINX (in London, UK).
In addition, VisualK highlights unusual behaviour. Flashing arrows show
load migrations between instances that are not considered adjacent. Root
name server instances start to blink if their traffic load decreases
significantly. The goal is to help spot unexpected changes while they are
happening and to allow root name server operators to investigate what
causes these changes.
For more information, please refer to the background article on RIPE
— Monitoring K-root in Near Real
*VisualK has been developed by Claudio Squarcella, intern at the RIPE NCC,
in collaboration with the Compunet
Roma Tre University.
*By Daniel Karrenberg <http://www.circleid.com/members/3167/>, Chief
Scientist at the RIPE NCC*
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the AfrICANN