44net-request(a)hamradio.ucsd.edu wrote:
Subject:
Re: [44net] Performance of DNS
From:
Brian Kantor <Brian(a)UCSD.Edu>
Date:
08/05/2014 07:21 PM
To:
AMPRNet working group <44net(a)hamradio.ucsd.edu>
On Tue, Aug 05, 2014 at 07:14:12PM +0200, Rob Janssen wrote:
>I often experience relatively slow lookups of
DNS records in .ampr.org and 44.in-addr.arpa.
It can be instructive to use the
'dig' '+trace' option to do lookups
as that will give you timing results as the query descends the tree.
That way you can get an idea of where the delay may be. Together with
the '@' option to direct your query to a particular nameserver you might
be able to identify the bottleneck when it occurs.
- Brian
I did some testing and I find that the two servers closest to me (in DE and UK)
return
results very quickly, under 80ms, while munnari.OZ.AU is very slow, it takes a second per
query.
Of course it is on the other end of the world, the pingtime is 350ms.
The lookup of org and ampr (when not in cache) also take 300ms each, so in total
a lookup takes quite some time.
When I trick the whole thing using these bind9 zones in my local caching resolver:
zone "ampr.org" IN {
type forward;
forward first;
forwarders { 192.109.42.4; 195.66.148.101; };
};
zone "44.in-addr.arpa" IN {
type forward;
forward first;
forwarders { 192.109.42.4; 195.66.148.101; };
};
everything is very very fast. of course this is to be expected, as the tree lookups
are no longer required and the fastest (for me) servers are used first.
But of course it is a dirty trick, and it will fail when those servers change address.
It looks like bind does not remember performance of DNS servers as it does for
forwarders,
or when it does it may have forgotten that info by the time it is required again and
therefore
does not use only the fastest servers?
Rob