Is Forwarding Faster?

[Mark_Andrews at isc.org]

> 
> I have wondered whether forwarding DNS queries from my local DNS
> nameserver is faster than having my local nameserver resolve them
> itself.  I reasoned that my ISP's nameserver would have quicker access
> to the Internet than my local nameserver.
> 
> I conducted a series of experiments to compare the amount of traffic
> generated by my local nameserver (Bind 8.2p1) when it was configured
> to forward queries with the amount of traffic it generated when it was
> configured not to forward queries. I used tcpdump to monitor IP
> domain-port traffic moving between my nameserver host and my ADSL
> modem.  I inserted or removed a "forwarders {<ISP DNS1>;<ISP DNS2>;};"
> statement in Bind's options block.  I killed and restarted named each
> time I inserted or removed the forwarders statment so as to begin each
> experiment with the same cache state, and I waited 60 to 90 seconds
> for named to initialize its cache before running dig.  I tried
> approximately thirty different domain names.  Once I had successfully
> looked up a domain name, I did not try it again until I had killed and
> restarted named.  For domain names I used domain names contained in
> URLs given by Alta Vista in response to an arbitrary search.
> 
> I found that when the forwarders statement was present, using dig to
> look up a domain name produced a single two-Ethernet-packet exchange
> between my local nameserver and the ISP's nameserver.  On the other
> hand, when the forwarders statement was absent, anywhere from four to
> more than 40 packets were exchanged between my local nameserver and
> various remote nameservers. 
> 
> I also observed the total query times that dig reported.  I found that
> when the forwarders statement was present, dig reported total query
> times that were almost always less than half a second and usually less
> than a quarter second.  When the forwarders statement was absent, dig
> frequently reported total query times of around one second.  In
> several extreme cases it reported times of five, six, and eight
> seconds, during which my local nameserver awaited responses to its
> queries.
> 
> The results suggest that even with an ADSL connection, a
> non-forwarding local nameserver is not as quick as my ISP's
> nameserver.
> 
> Payne Freret
> 
> P90/RH 4.2 Linux/Bind 8.2p1/PacBell ADSL
> 
> 
> 

	This is an un-fare test.  The fare test is to stop your
	ISP's nameservers as well.  You are not measuring what you
	say you are measuring.

	What you are doing is testing a nameserver with no significant
	cache against one with a large cache.  You would expect
	the one with the large cache to do significently better.
	The real difference will be the amount of time it take for
	the packets to transit the ADSL line (both ways).

	Mark
--
Mark Andrews, Internet Software Consortium
1 Seymour St., Dundas Valley, NSW 2117, Australia
PHONE: +61 2 9871 4742                 INTERNET: marka at isc.org