It might be worth looking at whether the servers support multiple queries over one connection, and whether they support pipelining, and what - if any - difference that makes. A 2nd, and any subsequent, query on a TCP connection won't incur the cost of a handshake, and if pipelining is handled correctly - using Nagling, or TCP_CORK/TCP_NOPUSH, then responses can co-habit TCP packets. I can tell from real-world metrics that at least some resolvers do support, and actually make, multiple requests over the same TCP connection - though they are rare. On Wed, Oct 17, 2012 at 1:49 AM, Anand Buddhdev <anandb@ripe.net> wrote:

[Apologies for duplicates]

Dear colleagues,

The RIPE NCC has just published an article on RIPE Labs, comparing the TCP and UDP response times of DNS queries to the root name servers:

https://labs.ripe.net/Members/bwijnen/tcp-udp-dns-soa-rt-ratio

Your comments, questions and suggestions are welcome.

Regards,

Anand Buddhdev RIPE NCC

-- Colm

Anand, On Wednesday, 2012-10-17 10:49:03 +0200, Anand Buddhdev <anandb@ripe.net> wrote:

The RIPE NCC has just published an article on RIPE Labs, comparing the TCP and UDP response times of DNS queries to the root name servers:

https://labs.ripe.net/Members/bwijnen/tcp-udp-dns-soa-rt-ratio

Just curious, was there any difference in IPv6 vs. IPv4 ratios? (In principle there should not be, however the IPv6 sockets API is a bit richer so maybe NSD had problems in IPv4 that it did not have in IPv6, for example.) Cheers, -- Shane