Re: On the use of multiple PA prefixes or a single PI prefix for IPv6 multihoming

On 22-sep-04, at 15:23, Cedric de Launois wrote:

> CONCLUSIONS:

> These observations show that, from a performance point of view, IPv6
> multihomed stubs get benefits from the use of multiple PA prefixes and
> should use them instead of a single PI prefix as in IPv4 today.
> This study thus strongly encourages the IETF to pursue the development
> of IPv6 multihoming solutions relying on the use of multiple PA
> prefixes.
> The use of such prefixes reduces the size of the BGP routing tables, 
> but
> also provides lower delays, more diverse Internet paths, which in turn
> yields to better possiblities to balance the traffic load and to 
> support
> quality of service.

"When considering dual-homed IPv6stubs, we see in fig. 13 or in fig. 14 
that the path diversity observed is already as good as the path 
diversity of a 25-homed IPv4 stub."

:-)

But the question is: will this help us or hurt us? Basically, BGP 
doesn't really know what the best path is, but it can usually detect 
and avoid the bad ones. So you pretty much always get something 
reasonable. With multi-address multihoming you get many more paths, 
some of which are better than what BGP would have given you, but a lot 
are worse. So the trick is to select the best one, or at least avoid 
the bad ones. How do we do this?

Iljitsch