Re: I-D ACTION:draft-ietf-shim6-applicability-01.txt

[marcelo bagnulo braun <marcelo@it.uc3m.es>]

El 16/06/2006, a las 20:18, Deguang Le escribió:
> According to your above definition, a host that is within a multihmed 
> site has also multiple network interfaces. From this point of view, 
> this situation descirbed in shim6 could also be regarded as host 
> multihoming. right? :-)

i guess it is mostly a matter of how to define a multihomed host

we could say that we have two types of multihomed hosts:

- the "strict" multihomed host where the host has multiple physical 
interfaces
- the "loose" multihomed host, where the host has multiple addresses

In the first case it is pretty clear that the host multiple paths to 
connect to the internet, since the attachment point is different.
In the second case, it may have multiple paths, not in the first hop, 
but afterwards, like the case of multiple ISPs assigning different 
prefixes to the multihomed site and using different addresses result in 
using different site exit paths. However, it should be noted it may be 
possible that having multiple addresses does not imply having multiple 
paths to the destiantions, (think the case of multiple RFC3041 
addresses)

So, i guess that the definition of multihomed site is kind of open, and 
i think this provides a good flexibility when using the term, but the 
cost is that some confusions may arise sometimes... :-(

> > There are aspects of host multihoming which are not part of the shim6 
> >  specification (e.g. the choice of egress interface and next-hop  
> > router that the host uses for a particular outbound datagram). In  
> > this sense, shim6 as it stands provides an inadequate specification  
> > for host multihoming.
> I think these issues that you point out(e.g. the choice of egress 
> interface and next-hop  router that the host uses for a particular 
> outbound datagram) should also be considered in shim6 since shim6 is a 
> host-based approach. right? :-)

not really, since the host assumes that using a different path will 
imply using a different ISP (not a different next hop, since there may 
be intermediate routers in the site)

regards, marcelo


> Cheers,
> Deguang
>
>
> > It seems to me that, for this reason, "site multihoming" is a better  
> > phrase than "host multihoming". However, given the host-centric  
> > approach that shim6 takes to site multihoming, additional text to  
> > this effect would quite possibly lend clarity to the document in  
> > question.
> > Thoughts?
> > Joe