I'll not continue with the charter topic, because you're right that it's
offtopic on a privacy list. However the point was made because it was an
opportunity as a protocol change would be introduced when introducing
encryption.

On Tue, Jan 17, 2017 at 09:49:09PM -0500, Robert Edmonds wrote:
> What does poor message decoding performance from a particular
> implementation have to do with DNS transaction confidentiality?
> 
> (There are DNS implementations that advertise high query rates with
> filtering, so I'm skeptical that backwards incompatible changes to the
> message format would deliver substantial speedups.)

No software implementations that I know of advertise link speed rates,
even considering a single 10gbe link. The fastest I've come across is
about 2.4 million QPS from a well-known commercial product. There's a
higher overhead to parsing DNS messages and constructing DNS messages[1]
than need be, especially when the service is hit with attack traffic.

The points I'd listed are from real situations we've encountered.

[1] Unlike for protocols such as HTTP that can serve pre-compressed
resources, DNS name compression is performed on-the-fly during message
rendering. It is a burden on implementations because efficient
implementation of matching would require implementing it in assembly,
something that a general purpose compression library can easily provide
if it were not DNS specific. I don't think any high-level language
implementation (C and above) can efficiently render something like a
COM. delegation (which contains 29 RRs) without a hit on
performance. Profile it and you'll see. If you argue that "DNS
implementations .. advertise high query rates", they could be doing
better.

                Mukund

Attachment: signature.asc
Description: PGP signature

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

Reply via email to