a little thought on exchanging traffic

Pickett, David dpickett at northc.com
Wed May 20 13:32:06 UTC 1998


> The fundamental problem is there are no magic pixie dust in this
business.
> Sure, some people like to put out press releases saying how they've
solved
> all the worlds problems using the Magic Frambulator.  But what they've
usually
> done is ignored half the problem. 

So, just how would the members of this group characterize this problem.
I am among the people who are "putting a lot of thought into this", and
would like a fresh re-hash of the issues.

> Or...they have designed systems that have the wrong goals...

> For example, some people design routers to "attract packets"
> like magnets. They design protocols for routers to tell other
> routers what packets to send them. Why on earth do people
> and companies want packets ? Processing packets costs
> time and money. Imagine a world where routers tell other
> routers what NOT to send them.

Curious angle.  Have route tables near the core of the network
gotten so large that it would be easier to advertise what NOT
to send?  I am dubious.

The fundamental job of an L3 forwarding agent is to apply a
forwarding policy to traffic that flows through it.  Whether
this policy is applied as positive or negative logic makes no
difference in terms of forwarding performance.  By advertising
which packets you DO want, you are implicitly telling a forwarding
engine which packets you DON'T want at the same time.

> Routers should be designed to "repel packets" and to quickly
> get rid of the ones they have.

By telling your neighbors that "I only want packets that meet
this specific criteria" are you not actively repelling all of this
packets that don't meet the criteria?

>                                 They should also be designed
> to send them as quickly as possible to the place they belong
> and not to some black-hole called a NAP so that people can
> puff out their chests about their NAP being bigger than the
> next NAP.

I doubt that anyone would take issue with your requirement that L3
forwarding agents, forward traffic as quickly and accurately as
possible.  In the same breath, you address the highly complex and
very different issue about how to interconnect these L3 devices.

L3 Forwarding devices implement policy.  The policies, in their most
basic form, tell the forwarding agents where, when, and how to handle
various classes of traffic.  What happens when competitive entities 
need to interconnect their L3 devices in order to build a larger 
network?  Does the current NAP model work well?  Do peering
agreements, as we understand them today, work and scale well? 

Thoughts?
dp

David R. Pickett
Northchurch Communications Incorporated
5 Corporate Drive
Andover, MA 01810
978-691-4649



More information about the NANOG mailing list