Re: /19 addresses and redundancy

[Phil Howard <phil () charon milepost com>]

Sean M. Doran writes...

[re: inbound filtering]

> Sprintlink did at one point.  It's a really good idea to do
> this in general because it mitigates the disconnectivity
> customers assigned prefixes out of one's address blocks
> will suffer if and when someone accidentally(?) announces
> subnet of those blocks.

Good point.


> Inbound filters can be adjusted, you know.  Unfortunately
> the people who have inbound filters have never figured out
> that they should make this a service that they charge for.

How easy is this to do.  How many filters would a company like
MCI/WorldCom have to place in each peer router?


> However, since inbound announcement filtering is a game
> anyone can play, I recommend people consider the
> implications of fee-based filter updating and how it can
> effect their routing whether or not they are the ones
> doing the inbound filtering.

Should the charge be for adding the filter (deny) or for deleting
it (permit)?

If the provider defaults to deny, then for a customer to request
a permit on their prefix means adding a filter to the router.
If the provider defaults to permit, then a customer wanting a
deny would obviously mean adding a filter.

Once a provider starts charging for the service under one policy,
that means that one set of customers pay the fee and the others
do not.  But if the provider decides to change the default, then
it reverses the customer sets of who pays and who does not.

One business model would be to choose the default based on the
largest number of customers paying.  But that could result in
customers leaving, either due to the extra fee, or due to the
slower operation of the network with so many filters in place.

The other business model would be the reverse, to choose the
default to minimize the number of filters, minimizing the costs
to the customers and maximizing the performance (while retaining
customer preferred route security policy).

I would tend to prefer the latter model.  I doubt such charges
could really make or break the bottom line for most businesses.
The tough position would be if the customer preferences went
about 50/50.

...

I still wish there was an easy way to filter routes on the
basis of allowing N route prefixes per prefix size per AS
where N might well be 1.


> Connectivity = bidirectional bandwidth + bidirectional reachability.
>
> Connectivity = value.

How might symmetry fit into that?

-- 
Phil Howard | end7ads0 () no61ads4 com blow6me1 () no3place edu stop1878 () anywhere edu
  phil      | w1x4y4z0 () spam3mer com no2spam6 () noplace9 org no8spam8 () spammer9 edu
    at      | ads2suck () no7place net a6b0c7d4 () no0where net die0spam () noplace0 com
  milepost  | stop9610 () dumb1ads edu eat0this () anyplace edu crash946 () noplace6 net
    dot     | eat4this () s0p7a8m7 com stop8it5 () dumb2ads edu ads2suck () s4p9a8m0 net
  com       | eat4this () spam8mer com no0way44 () anyplace net eat37me5 () no1place com