Re: "from the real world" - typical multihoming discussion

[Masataka Ohta <mohta@necom830.hpcl.titech.ac.jp>]

Sean;

> > It will be great if small operators do not have to use BGP merely
> > because they are multihomed.
> 
> Yes, agreed.
> 
> At the risk of summoning down fire from heaven, I would
> like to point out that NAT makes this sort of multihoming
> possible, and despite the often-alleged huge cost of NAT,
> some sites consider it easier to manage than BGP,
> particularly when it comes to managing the attraction of
> return traffic from sources across the Internet.

AFAIK, NAT based multihomed intranet (for which IETF is not a proper
standardization body) lacks rubustness.

Worse, the real cost of NAT is performance incapable of supporting
the broadband Internet.

> This is in large part because in general, BGP: is
> _difficult_; and is necessary to use merely because one is
> multihomed.

A good news is that proposals disabling small ISPs use global
routing table entries will unlikely to allow the ISPs have
their own ASes.

> One goal of the post-CIDR v6 IDR should clearly be to make
> multihoming maximally simple, on the assumption that:
> 
>         a/ nearly _everyone_ will want to do it & may have
>            the opportunity in the future
> 
>         b/ multihoming may happen for reasons of:
>                 i/   redundancy
>                 ii/  volume management (load balancing)
>                 iii/ performance management (use provider
>                         X as a return path from some places;
>                         use provider Y as an outgoing path
>                         towards some places, and so on)
>                 iv/  price management ("10+ dialling equivlanet")
> 
> CIDR does not make any of (b) easy or inexpensive, and 
> (a) will cause global routing to explode in the absence of 
> magical network renumbering technology.

>                 ii/  volume management (load balancing)

might be easy, but I'm afraid:

>                 iii/ performance management (use provider
>                         X as a return path from some places;
>                         use provider Y as an outgoing path
>                         towards some places, and so on)

might mean BGP or something as difficult as it.

							Masataka Ohta