[Pdns-users] Problems With Hosted DNS For Domain.

[Lorens Kockum]

On Sat, Sep 04, 2004 at 05:50:33PM +0100, James Napier wrote:
> 
> http://www.dnsstuff.com/tools/dnstime.ch?name=dsprofile.biz&type=A
> 
> Is there anything i can do about this in powerdns? Any
> extra records that need adding?

Don't worry about it. To begin with, as the site itself writes,
the tests are made from the US, so people in the UK will
have better response times, on the order of 60 to 70 ms at
least. Further, unless you're really *really* concerned about
response time and willing to spend money improving it, getting
the reply to your question in an average of under half a second
isn't that bad.

If you want to improve the "user experience" of your web site
visitors, remember that the DNS lookup delay is only for the
first access or every $TTL seconds. It's much more important
that the site load quickly, which depends on your bandwidth and
the client's bandwidth in relation to the size of your pages,
the load on your server, how dynamic your pages are . . . and
generally the design is important too :-) (Hint: an "under
construction" page is better than a page which asks you to
*define* the admin username and password :-) ).

For DNS servers, both servers responding is a really good start,
and even if you only have two and both are in the UK, they seem
to be very reasonably seperate and well-connected. As for the
time thay take to respond to questions (which is the only thing
that really concerns PDNS) they seem plenty fast enough, I get
response times ranging from 120 ms down to 40 ms depending on
where I query from, which when compared to tcptraceroutes[1]
indicate that when the reply is in the pdns internal cache, pdns
is generating a UDP reply in just three to five milliseconds
more than it takes the kernel to return a TCP ACK.

When the answer is not in pdns internal cache, the reply seems
some 300 to 400 ms slower, but that delay is difficult to pin on
your server from a distance (lots of route caching in routers).

HTH.

[1] Do check that you haven't turned off PMTU negotiation along
with the rest of ICMP.

-- 
#include <std_disclaim.h>                          Lorens Kockum