Re: [v6ops] Routers are hosts too!

[Joe Touch <[e-mail]>]

> On Jan 9, 2016, at 5:04 AM, t.petch <[e-mail]> wrote:
> 
> ----- Original Message -----
> From: "Joe Touch" <[e-mail]>
> To: "Alexandre Petrescu" <[e-mail]>;
> <[e-mail]>
> Sent: Saturday, January 09, 2016 6:26 AM
> 
>>> On 1/8/2016 8:05 AM, Alexandre Petrescu wrote:
>>> RFC1122 "Reqs for Internet Hosts -- Communication Layers" does not
>>> mention longest-prefix match.
>> 
>> Section 3.3.2.3 (source address selection) refers to Section 3.3.1,
>> which includes a set of mechanisms in Section 3.3.1.2 for selecting
>> among multiple gateways.
>> 
>> Note that this algorithm involves using a route cache with ICMP
>> redirects that help the system learn the "best" (quote from the RFC)
>> next-hop. Those entries are described in terms of host routes, but
> there
>> is a discussion about aggregating those entries in 3.3.1.3.
>> 
>> The net result of that aggregation is effectively longest-prefix
> matching.
>> 
>> So you're strictly correct - RFC1122 does not require longest prefix,
>> but a large number of entries in the route cache is more effectively
>> handled as such.
> 
> But then RFC1122 does not mention classless addressing or Variable
> Length Subnet Masks -

That was introduced 200 RFCs later.  

> as I recall, they did not exist at that time.
> Rather, RFC1122 deals with 32 bit Host addresses and Classful Networks,
> ie 8 or 16 or 24 bit addresses, so while you might have more than one
> entry in the cache for a given destination, they will all be the same
> length so the concept of longest-prefix (and of course, the term prefix
> has no meaning at that date) match does not come into consideration.

You can still do longest match for classful routing.  

Where the gist aggregates these route chache entries and how is up to them, but the alternative is a nonscalable set of host routes 

Joe

> 
> Tom Petch
> 
>> Joe
>>