> Will the NVO3 dataplane protocols have mechanisms to detect loss, so
> they can actually do RFC4821?

That would definitely be worth calling out; I'll see about writing
some text ... in my "copious spare time" ;-).

> (I've often seen 4821 cited alongside 1981, because people think it's
> a drop-in replacement when it isn't.)

Some of the dataplane protocols under consideration do not have that
ability, and in addition, there's a strong tendency in nvo3 to shove
this problem up the stack, i.e., expect the applications using the
overlay to do their own MTU discovery.

Thanks,
--David

> -----Original Message-----
> From: Eggert, Lars [mailto:[email protected]]
> Sent: Thursday, May 15, 2014 3:38 AM
> To: Black, David
> Cc: [email protected]; [email protected]
> Subject: Re: [tsvwg] Fragmentation and Path MTU text in nvo3 dataplane reqts
> draft
> 
> Hi,
> 
> On 2014-5-14, at 22:52, Black, David <[email protected]> wrote:
> >         o Classical ICMP-based MTU Path Discovery [RFC1191] [RFC1981] or
> >           Extended MTU Path Discovery techniques such as defined in
> >           [RFC4821]
> 
> Will the NVO3 dataplane protocols have mechanisms to detect loss, so they can
> actually do RFC4821?
> 
> (I've often seen 4821 cited alongside 1981, because people think it's a drop-
> in replacement when it isn't.)
> 
> Lars

Reply via email to