[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
On Mon, 24 Jun 2002, Dave Wilson wrote:
> Imagine network A does not (under current RIR rules) qualify for a /35.
> Network A has routes to network B via (a) a high speed direct connection,
> (b) a high-speed network and (c) an ordinary commodity internet connection.
> Network A provides connectivity to a number of sites (< 200) not under its
> direct control.
> In IPv4, where Network A has an ASN and allocation from a RIR, BGP localprefs
> enforce a routing policy which uses the high-speed connects in preference to
> the commodity internet.
Only for outgoing traffic. Balancing incoming traffic is much harder,
unless the networks sending the traffic cooperate.
> Does section 3.1.4 require a solution to this scenario? "class of traffic"
> might just mean a protocol or could be wider, so I am not sure.
> The reason I bring this up is that it is not an unusual configuration for a
> national research network. If the answer is "no" then it is not necessarily
> a problem with the document :) but if the group [does|does not] plan to
> address it, I think it would be useful feedback for the policy groups.
A multihoming solution that doesn't provide for a way to make a
higher-bandwidth connection to be more preferred than a lower-bandwidth
connection won't fly, because it doesn't address real-world needs. There
are still many places in the world where bandwidth is expensive enough
wasting a significant amount of it is not an option.
Route selection and load balancing are mainly a problem with solutions
that require hosts to make decisions that ultimately determine routing.
(Ie. where a host must select one address out of several to communicate
with.) At some point, routing and routing policy information has to reach
those hosts. Since this doesn't exist yet, we are free to introduce all
the features we think are necessary.
However, this requires multi6 to start looking at solutions. IPv6 adoption
doesn't seem to happen at a break-neck pace, but at some point operational
needs will make multihoming happen one way or another. (How long will the
RIRs refrain from giving themselves provider independent address space,
for instance?) That genie will be hard to put back in the bottle.
Iljitsch van Beijnum