Re: RE: AD request / L2 Triggers Chapter Statement

Phil Neumiller wrote:

>>But there isn't; IP currently won't do anything interesting with info on
> 
> Right, but many routing systems are built ON IP and below transport.
> Take MIP, any MANET protocol, ICMP itself, OSPF, BGP4, etc.  All of
> these systems that reside at the IP layer in routers CAN INDEED make
> extremely good use of this information.

Which of these operates on millisecond decisions?


>>>  Applications include but are not limited to:
>>>
>>>o      Fault tolerant failover of links
 >>
>>Usually within a single link layer type - i.e., from one 300Mhz wireless

> This is where you are completely missing it.  The whole point is for routers
> with several heterogeneous wireless links to policy route over them in the
> event of failure.  I think you are thinking in terms of a simple host.  Think in
> terms of what a MANET or router gains from this.

What I'm asserting is:

	if it's within two links of the same type, a meta-link can do it

	if it's between two different links, IP routing has to do it,
	and that doesn't care about millisecond events

>>>o      L2 signal quality indications
>>
>>To what end? What would IP do differently?
> 
> It would most likely simply convey this to the transport layers that have
> registered interest.  It may do this with an ICMP message.

Name a transport layer that reacts on millisecond timescales.


>>>o      L2 congestion control
>>
>>Again, to what end? IP doesn't have congestion control; TCP doesn't 
>>react on milliseconds. 
> 
> TCP DOES react in milliseconds on the boxes (routers) and hosts I work
> on.  It will get even faster as time goes on.

TCP reacts on RTTs, usually multiple RTTs. Even then, what would TCP do 
that it doesn't already to? I.e., you don't need an L2 trigger to 
program a box to send TCP ECN data back.

>>>o      L2 fragmentation control
>>
>>There's an existing mechanism for that - MTU. Does it change on the link
>>
>>on the fly?
> 
> In many instances it would be very NICE to have the option.  QoS queueing
> mechanisms can benefit greatly from such capabilities.

Again, if you want to do this on a single box, you don't need an L2 
trigger. It's not that you're talking to IP in general, but that you're 
talking to a particular stack implementation.

Joe






_______________________________________________
pilc mailing list
[email protected]
https://www1.ietf.org/mailman/listinfo/pilc
http://www.ietf.org/html.charters/pilc-charter.html
http://pilc.grc.nasa.gov/