Re: [v6ops] Routers are hosts too!

[Alexandre Petrescu <[e-mail]>]

Le 08/01/2016 19:15, Jeff McAdams a écrit :
>
> On Fri, January 8, 2016 13:00, Alexandre Petrescu wrote:
>> In that sense, I dont know what makes a Host be a Host and not a Router.
>
>
>> What particular characteristic of Host is not a characteristic of a
>> Router too?
>
> So, I'm still not sure what you're trying to accomplish out of this
> discussion, but ok....

The Host-Router distinction is omnipresent in RFCs.  Whenever writing 
text one has to be careful about it.  There are many RFCs exclusively 
dedicated to Routers, others exclusively dedicated to Hosts.

> I think you have it backwards.  At least in my mind and the working
> definitions I've always used, a router is a device that forwards packets
> between network interface using Layer 3 addressing information.  A device
> that *doesn't* forward traffic would thus *just* be a host.

This is a definition I can agree with.  There is more to it, however, 
especially when it comes to the presence of the rt table and algorithms.

> I've never worked up a good working definition in my mind to decide
> whether routers are hosts or not...mostly because, in 20+ years of my
> career, it has never been something I've had to even think about.  Thus my
> questioning of what this whole discussion is really all about.  We're - in
> my mind - well into the territory of angels dancing on the head of a pin,
> much ado about nothing, and a huge heaping dose of bikeshedding.
>
> If you find an example of a system that *doesn't* do longest match, then
> let's talk about it.  (Yes, I saw the example of uIP, I would argue it
> does do longest match as it chooses between localnet destinations and
> 0.0.0.0/0...it just has an *extremely* constrained routing table capacity
> - 2 entries, and a very unusual internal representation of that routing
> table)

It's maybe about finding such a system, or about writing one.

In an IP stack the operations needing most computing power are at 
longest-match and at the CRC computation.  Their elimination can lead to 
very performant stacks - for Hosts.

I mean stacks that must run in constrained CPUs and still react very fast.

It is maybe researchy angels on top of trees... or maybe go the right 
WGs needing it.

Alex

>