Re: Draft of updated WG charter

[Iljitsch van Beijnum <iljitsch@muada.com>]

On 10-jan-04, at 6:52, Jay Ford wrote:

> The hosts & associated end users are not up to the job of doing 
> anything
> resembling intelligent path selection.  A scheme in which the
> source/destination addresses selected by a host dictates a path seems
> unlikely to work well.

> The core routers know enough to make a good choice among the border 
> routers.

> The border routers know the correct next hop for all destinations.

I'm sorry, but I have to disagree. The only thing that border routers 
know, is  that _if_ something is reachable, what the next hop is. 
Whether or not something is indeed reachable is burried beneath layers 
of aggregation. Even if we can make the information for all sites 
available without aggregation (which I don't think we can, re the need 
for route flap dampening in the late 1990s when we already had 
aggregation), there is no way we can make routers keep track of the 
reachability status of individual hosts. However, it shouldn't be too 
hard to make hosts keep track of the reachability status of (the 
different addresses of) the hosts they're actively communicating with. 
TCP already does this in single-address fashion for the purpose of 
congestion control.

> This is a good split of complexity & responsibility, with the hosts not
> having to do any of it.

I'm sure this model can accommodate some situations, such as carefully 
managed networks. However, many networks aren't managed to any 
noticeable degree (home networks), or the network only consists of a 
number of hosts that happen to be in each other's vicinity for a while. 
In these cases, hosts must step up to the challenge.

It occurs to me that IPsec tunneling in VPN boxes vs running SSL on the 
end-hosts has many similarities to what we're discussing here. Is there 
anything we can learn here?