On Thu, Sep 17, 2020 at 08:50:33AM -0700, Paul Vixie wrote:
> Mark Andrews wrote on 2020-09-16 15:20:
> > ... O(10^6)
> > But DNS traffic doesn’t need those sized packets even for zone transfers.
Not today, but DNSSEC with post-quantum signatures might prove tough to
squeeze into even 1400 bytes... 10^6 sounds quite remote, but I'm
already seeing (just shy of) 64KB frames between the kernel and 40Gb
NICs that do TCP reassembly and segmentation in hardware.
> i feel the same. but i felt that way about 64K and 640K RAM, and was
> wildly wrong. in practical terms i think the old 4K EDNS bufsize default
> will likely remain relevant to DNS, but two things: first, there are
> other things on the internet besides DNS; second, an evolving packet
> size that scales with bit rate would and still can obviate some of the
> things we use fragmentation, or TCP, or shortly QUIC, to do. the tyranny
> of 1500 must end, for the sake of our collective future.
I agree, but getting there will not be easy. But indeed over-optimising
for the present is an impediment. Step 1 might be to actually agree on
the
--
Viktor.
_______________________________________________
dns-operations mailing list
[email protected]
https://lists.dns-oarc.net/mailman/listinfo/dns-operations
