Re: [DNSOP] bar bof, edns buffer size, avoid fragmentation

sounds like a delightful session with some productive ideas.

On Wed, Jul 24, 2019 at 5:23 PM Paul Vixie <paul@redbarn.org> wrote:

> one other matter emerged during this discussion. path mtu discovery is
> dead at
> the moment, for valid security reasons. however, a jumbogram sized campus
> can
> be described in static routes, using MTU 1500 only for "default". and
> also,
> any future path mtu discovery functions will, as in the past, simply
> create
> "host routes" (/32 or /128) to memorialize the MTU known for that host,
> since
> that is where the TCP MSS computation will expect to find it. so, the lack
> of
> path mtu discovery in today's networks does not invalidate the method of
> calculating TCP MSS from the route MTU; thus, this should be safe for EDNS
> buffer size as well.
>
> paul
>
> re:
>
> On Thursday, 25 July 2019 00:19:18 UTC Paul Vixie wrote:
> > with four of us in attendance monday evening, i presented my appreciation
> > for fujimora-san's draft and admitted that EDNS0 ought to have required
> DF
> > and ought to have used examples which would fit in a then-common MTU 1500
> > network.
> >
> > i then asked that numbers like 1220, 1280, 576, and 1500 not be
> specified as
> > fixed constants, but rather, should be computed in the same way and for
> the
> > same reasons as TCP MSS. those of us running jumbograms or still running
> > FDDI or perhaps speaking POS should not be artificially limited in our
> DNS
> > payload sizes.
> >
> > in discussion, the following details appeared:
> >
> > TCP MSS is calculated from the routing table. on microsoft windows,
> there's
> > an easy way to find the route for a destination...
> >
> >
> https://docs.microsoft.com/en-us/windows/win32/rras/search-for-the-best-rout
> > e
> >
> > ....and while there's no portable way to do this from user mode in Linux
> or
> > BSD or other UNIX-style systems such as Android, there is always a way.
> >
> > it is the route's MTU and not any one interface's MTU, and not any
> constant,
> > that should be used to calculate EDNS0 buffer size. that will probably be
> > 1280 or 1220 for most V6 routes (including "default", the route of last
> > resort) and 1460 for most V4 routes (including default).
> >
> > (noting, i've made a similar plea to the QUIC team. the 1500 MTU will not
> > last our lifetimes, and we must not hard code the 1500 MTU assumption...
> > anywhere.)
> >
> > also, there were beer and snacks.
>
>
> --
> Paul
>
>
> _______________________________________________
> DNSOP mailing list
> DNSOP@ietf.org
> https://www.ietf.org/mailman/listinfo/dnsop
>