HE.net BGP origin attribute rewriting

Sat Jun 2 11:45:19 UTC 2012

On 06/02/2012 12:43 PM, Joe Provo wrote:
> Last post on this topic for me. You seem to wish to argue 
> against the lessons of history and the reality of running
> a network on the global Internet.

Based on observations from routeviews / RIPE RIS / other public sources,
overwriting BGP origin isn't a common practice. I did some analysis
before I opened this topic.

>From tier-1 networks, only Level3 seems to do this, from other major
networks only HE. Based on network listed at
http://en.wikipedia.org/wiki/Tier_1_network, there're 2 of 22 major (and
only 11 tier-1) worldwide networks performing origin overwritting.

That's really not a representation of common and widely used practice.
I'm not arguing with common practice on the internet. Majority doesn't
touch origin attribute...

(and yes, basically I don't care about pure tier-2/3 networks, their
impact isn't peremptory in terms of their global impact)

> The two issues are orthogonal. Deaggregating sources have 
> been cost-shifting [in a highly visible and easily examined
> and often trivially-filtered] manner for ages.

In global table, there's 41% overhead, in terms of prefixes announced. I
don't think it's trivial to filter this overhead. If you're correct (I
don't think so), why there's this huge ammount of unfiltered
deaggregated prefixes in global table? Because it's easier to buy new
hardware.

> A midspan network deaggregating someone else's prefixes is 
> broken and gets called out, generally by the originator if 
> they have a clue.

This is bad at all - but sometimes also happens with huge impact and
this is historically documented on some cases like Pakistan
Telecom/Youtube. And this happened, even you said that filtering is
trivial...

- Daniel