Re: [Int-area] Side meeting follow-up: What exact features do we want from the Internet? minmax

["Templin (US), Fred L" <[e-mail]>]

Sorry Alex, I spoke imprecisely. In ATM, there is one fixed cell size but for
heterogeneous Internetworks the 576 determines only a lower bound for cell
sizes - not an upper bound. In ATM the fixed cell size is 53 octets (5 octets header;
48 octets payload), but for heterogeneous Internetworks we need to leave a lot
of extra room for headers. So we define a "minimum Maximum Payload Size
(minMPS)" of 400 octets. That leaves a generous 176 octets for encapsulation
headers, which will most often not be fully consumed due to header compression.

Again, this defines only the minimum MPS; larger MPS values can be discovered
through path probes, or can be set optimistically under the risk of encountering
unexpected size restrictions.

Fred

> -----Original Message-----
> From: Int-area [mailto:[e-mail]] On Behalf Of Alexandre Petrescu
> Sent: Wednesday, December 08, 2021 6:16 AM
> To: [e-mail]
> Subject: Re: [Int-area] Side meeting follow-up: What exact features do we want from the Internet? minmax
> 
> The advantage of talking 'minimum cell size' (MCS?) instead of 'maximum
> transmission unit' (MTU) is in that it fits better to a preceding
> 'minimum' qualifier.
> 
> A 'minimum MTU' term expands to a 'minimum maximum transmission unit'.
> That construct can be ambiguous to an implementer.  Is it a 'minimum
> transmission unit'?  Or is it the smallest of all MTUs of all link designs?
> 
> On another hand, a 'minimum minimum cell size' is clearly just a minimum
> cell size, just further accentuated in its minimization: a 'minimum
> minimorum' if one wishes.  Where MCS is 576 a minimum MCS in IPv6 could
> be 41: 40 bytes of header and 1 byte of payload.  Or maybe 40 if there
> is no payload.  Or maybe 321 bits if there is just one bit of payload
> (an on/off switch or flag in IoT speak).
> 
> Alex
> 
> Le 08/12/2021 à 02:38, Templin (US), Fred L a écrit :
> > This conversation is missing some fundamental points – really the
> > most important
> >
> > points – which are the minimum sizes guaranteed to work everywhere.
> > For IPv6,
> >
> > the minimum MTU/MRU are 1280/1500. For IPv4, they are only 68/576 but
> > since
> >
> > the IPv4 network supports fragmentation we can nominally designate
> > the IPv4
> >
> > minimum MTU as 576 also if we clear the DF bit. It means that,
> > without probing
> >
> > or having some divine knowledge of paths that have not been
> > previously visited,
> >
> > the ONLY sizes guaranteed to work are 1280 for IPv6 and 576 for
> > IPv4.
> >
> > Now take the case of Multinet where a path may traverse multiple
> > concatenated
> >
> > IP networks of arbitrary IP protocol versions - remember “Catenet”?
> > Since there
> >
> > may be no advanced knowledge of network IP protocol versions, the
> > most we can
> >
> > absolutely and for sure count on across the entire path is 576.
> >
> > What this gives us is not the **maximum packet size**; instead, it
> > determines the
> >
> > **minimum cell size**. We know that a 576 cell will traverse all
> > paths, so we never
> >
> > send a non-final cell smaller than this (which might trigger a tiny
> > fragment alarm).
> >
> > But, we can certainly send packets larger than the path MTU -
> > **much** larger in
> >
> > many cases. And for paths that support them, we can also send Brian’s
> > jumbograms.
> >
> > Speaking of jumbos, on a different list I made a point about how this
> > all harkens back
> >
> > to RFC1149 and RFC6214. But nowadays, we can imagine substituting a
> > memory stick
> >
> > for the slips of paper with whitestuff and blackstuff and those avian
> > carriers will
> >
> > transport jumbos just fine. Not quite unlimited MTU, but close
> > enough.
> >
> > Fred
> >
> >
> > _______________________________________________ Int-area mailing
> > list [e-mail]
> > https://www.ietf.org/mailman/listinfo/int-area
> 
> _______________________________________________
> Int-area mailing list
> [e-mail]
> https://www.ietf.org/mailman/listinfo/int-area