Re: Source address selection in IPv6 multihomed multi-addressed sites

[Xiaowei Yang <yxw@cordelia.lcs.mit.edu>]

> 
> * A sixth solution is that a host tries to detect
>    by itself which source address to use. A proposition
>    is that a host maintain by itself one or more full BGP
>    tables so that it can deduce the source address to use.
> 
>    Pro : no tunnel
>    Cons: - waste of computer resources (how much ?)
>          - no easy configuration

I do not see why a host needs to maintain the full BGP tables.  For
source address selection, it seems to me a host only needs to know the
provider-level route a source address represents, i.e., the chain of
providers that allocate a source address to the site/host. Suppose the
source address allocation chain on average has a depth of three. A
network on average takes source addresses from five providers. Then
the number of providers concerning a site's source addresses is about
5^3.  If a graph is used to represent these providers, the resources
required to store this graph and update it should be much less than
keeping an updated full BGP table that has 100,000+ entries.

The destination address selection is also a problem. To send the first
packet to a multi-homed destination, a host needs to pick a valid
destination address. This picking is less important because once the
first packet reaches the destination, the destination host may switch
to a different address in its reply packets that fits its QoS
requirements. Therefore, a host may random pick a destination
address. If it doesnot work, the host could try the a different
destination address.

Obtaining the provider information concerning source addresses of a
host requires a new network service.  Source address selection implies
provider selection, which has the potential to increase competition
among providers and discipline the ISP market. From this aspect, it is
important for a user to make an informed choice rather than a random
choice. Thus, the effort to provide a new network service to
disseminate provider information may be worthwhile.

My thesis work is right on this topic. If anyone is interested, a
short paper describing my work can be found at
http://www.isi.edu/newarch/DOCUMENTS/yang.nira.pdf

Xiaowei