Re: [core] draft-tschofenig-core-coap-tcp-tls-03 : TCP Session Establishment

["weigengyu" <[e-mail]>]

Grüße, Carsten wrote:
> (Talking about the UDP case:)
> That is indeed an area where the current IPv6 architecture doesn't
> provide as much help as one would like it to.  The MTU is not
> necessarily a good choice, in particular when hops with adaptation layer
> fragmentation are involved.

I agree.
MTU just limits the largest size of PDU.
And it is short of standards to tell what is the fittest and most efficient.

To get efficient tranfer within a constrained network,
it is likely to get one application message into one data link frame.
But, the application has no way to have this cross layer information.

Maybe, it is required to define something as draft-bormann-intarea-alfi-04.

Regards,

Gengyu WEI
Network Technology Center
School of Computer
Beijing University of Posts and Telecommunications
-----原始邮件----- 
From: Carsten Bormann
Sent: Thursday, April 23, 2015 2:55 PM
To: weigengyu
Cc: Carey, Timothy (Timothy) ; [e-mail]
Subject: Re: [core] draft-tschofenig-core-coap-tcp-tls-03 : TCP Session 
Establishment

weigengyu wrote:
> But what the application depends on to make the decision?
> From the viewpoint of network, it should be MTU.
> The MTU limits the size of UDP and CoAP message.

(Talking about the UDP case:)
That is indeed an area where the current IPv6 architecture doesn't
provide as much help as one would like it to.  The MTU is not
necessarily a good choice, in particular when hops with adaptation layer
fragmentation are involved.

The problem is mitigated a bit by the fact that usually a constrained
network with a preference for small packets is the last (incoming) or
first (outgoing) hop to a constrained node, so this node may have more
insight into good packet size choices than the network as a whole is
able to provide.  Using -block, this node can then communicate its
choice to a peer.

A while ago, I have proposed a way to get more (and more detailed)
information out of the network:
https://tools.ietf.org/html/draft-bormann-intarea-alfi-04
Earlier drafts of that had the problem that IPv6 hop-by-hop options tend
not to traverse the general Internet intact.  So I came up with a
SPUD-like alternative solution in -04 (see its section 5:
https://tools.ietf.org/html/draft-bormann-intarea-alfi-04#section-5
).  With (the real) SPUD now being looked at for communication to
middleboxes in general, that basic approach might actually go forward later.

Grüße, Carsten