Filter-based routing table management (was: Re: minimum IPv6 announcement size)

Scott Brim scott.brim at gmail.com
Thu Sep 26 21:05:02 UTC 2013


Oh this sure will be fun. For a good time, see how GSMA handles
connectivity with IPXs.
On Sep 26, 2013 1:28 PM, "William Herrin" <bill at herrin.us> wrote:

> On Thu, Sep 26, 2013 at 11:07 AM, John Curran <jcurran at istaff.org> wrote:
> > On Sep 26, 2013, at 4:52 AM, bmanning at vacation.karoshi.com wrote:
> >
> >> sounds just like folks in 1985, talking about IPv4...
> >
> > If there were ever were a need for an market/settlement model, it is
> with respect
> > to routing table slots.
> > That's not to say that establishing a framework for externalizing
> routing costs would
> > be easy; it's a complicated and twisted matter, and also fraught with
> various legal &
> > competitive aspects.
>
> Hi John,
>
> That's putting it mildly. Establishing such a framework would be an
> immense challenge. Here are some ideas I've heard:
>
>
> 1. The International Clearinghouse
>
> Every BGP participant files with a clearinghouse, specifying:
>
> a. How much they charge to carry 1 route
> b. Whether or not they are a leaf node
> c. Whether or not they are a transit-free network.
>
> Any network which is not transit free must implement a default route
> which leads to a big transit-free network in order to maintain full
> connectivity.
>
> The BGP participants then publish the exact routes they intend to
> announce to the clearinghouse and for each one select which networks
> they'll pay to carry the route. The route must still reach each
> network via BGP; payment just means that the network won't filter the
> route out.
>
> The clearinghouse then collects payments from everybody and makes
> payments to everybody, as well as providing each participant a list of
> the routes that are paid for. Sellers are expected to promptly
> incorporate new paid routes into their BGP filters.
>
> From my research a few years ago, a reasonable rate would be around 3
> to 4 cents per year per advertised route per BGP-carrying router in
> the organization. A couple billion dollars per year if the routing
> table maintained its current size.
>
>
> 2. The partial routing scenario
>
> Large service providers put bids in to the RIRs for the right to
> announce /8 covering routes for each /8 delegated to the RIR. Each /8
> matches exactly one service provider. Smaller BGP system participants
> make private arrangements with a small (20 to 30) set of networks
> (including their direct ISPs) to carry their advertised routes through
> a reasonably redundant number of pathways to (and including) the
> winning bidder for the /8 they inhabit. For the sake of performance,
> they may also pay additional large networks to shortcut the traffic
> towards them rather than let it dump at the /8 advertiser.
>
> For the folks you don't pay via the clearinghouse, many end-user
> systems and the majority of transit systems simply don't carry your
> route unless yours is among the handful of systems critically
> important to their customers. Instead, traffic to your network follows
> the /8 advertisement until it reaches a network which carries your
> specific route.
>
> With the routing costs suitably reduced, settlement for the remaining
> routes becomes moot.
>
> This is usually within a few percent of the routing efficiency that
> would have been achieved with total route propagation.
>
>
> 3. The routing overlay
>
> Establish a semi-stateless tunneling system. Each BGP participant sets
> up a tunnel ingress node and links a default route to it. Packets for
> a destination not found in the routing table follow the default route
> to the tunnel ingress.
>
> The tunnel device then looks up an tunnel exit node via a mapping
> protocol. Both the map server and the exit node have to be hosted on
> IP addresses reachable via the normal routing table.
>
> Having found an exit node, the original packet is encapsulated into a
> tunnel packet and sent to the exit node. The exit node is in a part of
> the network that carries an explicit route to the destination.
>
> Then, move the definition of threshold size. Except for whitelisted
> critical infrastructure, /24 advertisements would no longer carry an
> expectation of universal distribution. To maintain connectivity, folks
> at the bottom of the chain would need to establish or subscribe to
> tunnel exit nodes that have a route back to them.
>
> With the routing costs suitably reduced, settlement for the remaining
> routes becomes moot.
>
> The IRTF Routing Research Group studied such protocols a few years ago
> and have pretty well fleshed out how to make one work with all the
> tangled issues involving path mtu, dead path detection and so on.
> Multiple designs sit on a shelf waiting for a promise that the
> technology will be purchased if built.
>
> Regards,
> Bill Herrin
>
>
> --
> William D. Herrin ................ herrin at dirtside.com  bill at herrin.us
> 3005 Crane Dr. ...................... Web: <http://bill.herrin.us/>
> Falls Church, VA 22042-3004
>
>



More information about the NANOG mailing list