Re: Full routes needed at sites? [Re: RIR bashing, was: Routing table size?]

[masataka ohta <mohta@necom830.hpcl.titech.ac.jp>]

Juan Rodriguez Hervella;

> > What end users CAN NOT accept is complex configuration.

> > A proposal to force end users some action is NO acceptable.

> So it should be better to let the end-user deal with the full IPv6
> routing table, I think he/she will have a lot of fun seeing a lot
> of prefixes on the screen when he/she can not reach some host.
> This is just a funny comment, cause I totally agree with you on this
> point.

That is

Pekka> It's possible to remove the default route for the other ISP if it's
Pekka> broken, or to add more specifics to point to the working ISP if
Pekka> necessary.

can not be an option for end users.

> > With BGP configuration effort of an end user is variable and,
> > if most end users can just use AS-path-len, it can be a default
> > of most routers.
>
>
> As Pekka told in a previous mail on this thread, I think 
> that making a BGP solution for SOHO users seems
> a bit overkilled.

Pekka eventually said, don't rely on routing protocols and use
static routing, which is the overkill.

Non-BGP solution for SOHO users is the overkill.

End users can swallow full routing table, if the size is
reasonable and inexpensive routers can accept by default.

Your example;

> mh-router-shell (config)> ipv6 multihomed up

is unnecessarily complex. That is, router comfiguration should
be same, regardless of whether you are multihomed or not and
the factory default should work in all cases.

> Besides, I would like to see what requirements do you need
> to build up your solution. Have you come up with some of them ?

I already suggested one:

> > With BGP configuration effort of an end user is variable and,
> > if most end users can just use AS-path-len, it can be a default
> > of most routers.

> I think that one of the problems (problems = requirements for end-users)
> of your solution is that you will have to have ASNs for every multihomed-user.
> And I guess this is only the beginning.

Not at all.

As the end users do not offer transit, they don't really have to
have ASNs. The end users accept all the route and generate none.

So, we can modify BGP. Or, keep BGP as is and have a psuedo ASN
shared by all the end users. BGP daemons on border routers of
ISPs should filter out any information from the end users that
the psuedo ASN won't be visible.

> I can not comment without further info. Will you make a requirements-draft 
> so we can start making some comments on it ?

The requirement is to make routing table reasonably small.

Have you read:

	draft-ohta-multihomed-isps-00.txt

> > > > How many of multihomed (non-transit) entities today are using the
> > > > default-only approach?
> > >
> > > Good point,
> >
> > It is phychologically unacceptable for operator of a multhomed
> > site that paid connection to some ISP will NEVER be used.

> "Money for nothing" is not a good deal, that's clear.
>
> Just I don't understand why you say "it will NEVER be used".
> It's our duty to develop mechanisms in a proper way so things get to
> be used, is not it ?

Oops, I mean "NEVER unless the default ISP is completly down".

I won't request (outgoing) load be fully distributed over
all the external links. But, if you are triply homed, and
if the third link is not used for outgoing traffic unless
the first and the second ISPs are down, it is not acceptable.

> > > though I don't think the IPv4-way of doing multihomed things
> > > should be the example to follow for IPv6-multihomed sites...
> >
> > which is what RIRs are doing.
>
>
> Omg, I'm at least I'm on my last comment.... *wink*
>
> I dunno if that's already been said, what should the RIRs be doing ?

We can't expect much on RIRs for protocol development.

						Masataka Ohta