[e2e] What should e2e protocols know about lower layers?

Joe Touch touch at ISI.EDU
Wed Oct 10 10:08:54 PDT 2001


RJ Atkinson wrote:

> At 10:34 10/10/01, Tim Moors wrote:
> 
>>One thread of the discussion questioned how an instance of an e2e 
>>protocol like TCP can determine whether its communicating peer is
>>"local".
>>
> 
> I do not think "local" can be determined above layer-2.  I can't 
> even convince myself that it is always something that can be
> determined at layer-2.  Depending on what "local" means, even
> a single Ethernet might not be "local" (e.g. 1 GigE products
> often have 70-100 Km reach now a days; most shipping GigE 
> products also support MTUs larger than IEEE-standard size).


I guess I'm the one on the other side of the debate.

IPv4 certainly has a notion of local - the subnets to which you are 
directly connected. This is why, IMO, broadcasts are permitted there.

IMO, a network that has a long link (as above) that is not configured as 
point-to-point is misconfigured.

As Ran observed on the TSVWG mailing list, IPv6 lacks broadcast, one 
could argue for reasons related to this. However, replacing the 
well-known term broadcast with the magical 'anycast' doesn't replace the 
need.

The LAN is an important component of the current architecture, and 
differentiates between local and nonlocal for the use of broadcast to 
avoid centralization of MAC address registration, and the alternative of 
explicit dispersed registration.

--

While I appreciate the use of congestion control throughout, I still 
maintain that there are times when the network layer in particular needs 
to distinguish between local and nonlocal. I see no reason why the 
transport layer shouldn't be afforded the same opportunity.

Joe





More information about the end2end-interest mailing list