Benno Overeinder <[email protected]> wrote:
>
> All solutions above are stable performant DNS-over-TLS implementations.

I think what Ian is referring to is support for concurrent queries and
out-of-order responses.

I briefly tested a few resolvers starting from an empty cache, making 2000
queries with a max concurrency of 100 outstanding queries. I'm using
`adns` which pipelines queries down one persistent TCP connection.

With BIND 9.11 (concurrent TCP queries) it took 42s over UDP and 42s over
TCP.

With BIND 9.10 (serialized TCP queries) it took 72s over UDP and over TCP
it took 427s - plus I had to reduce the TCP concurrency to 10 to stop the
client timing out, and even with this limited concurrency, a couple of
slow queries can cause several following queries to time out even if they
would otherwise have succeeded.

With Unbound over UDP this test completed in 110 seconds, but over TCP it
took 870 seconds, and again I had to restrict the concurrency to reduce
timeouts.

There's a lot of variation in the times because most of the delay is due
to broken authoritative servers (the 2000 queries come from a very crappy
dataset) and Unbound and BIND have different strategies for handling
brokenness.

Unlike my test, for DNS-over-TLS the delay we usually care about is the
stub/resolver latency (because much of the use case involves talking to
resolvers that are further away). It's important that the stub doesn't
have artificial concurrency limitations so it can compensate for this
latency without having to work around head-of-line blocking. It's also
important that the stub can use a persistent connection so it isn't paying
for the multi-round-trip setup cost on every query.

Tony.
-- 
f.anthony.n.finch  <[email protected]>  http://dotat.at/
Trafalgar: Cyclonic gale 7 to severe gale 9, occasionally storm 10 for a time
in northwest, becoming northwesterly 5 to 7 later. Very rough or high,
occasionally very high for a time in west, becoming rough or very rough later.
Rain or showers. Good, occasionally poor.

_______________________________________________
dns-privacy mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/dns-privacy

Reply via email to