🚀 go-pugleaf

On 13 Mar 1999 23:52:32 GMT, 
newhouse at rockhead dot com <newhouse@nospam.crap> wrote:
>In article <slrn7ej45a.6bf.ese002@spica.exile.org>,
>	ese002@news9.exile.org (Eric Edwards) writes:
>> It's a bit fuzzier for large ISP's with private backbones (like Concentric).
>> I expect backbones that wish to provide service for other ISP's to use 
>> private peering.
>
>It might be fuzzier for you. 

ISP1 -> private backbone -> NAP -> backbone2 -> ISP2

is clearly better than

ISP1 -> backbone1 -> private peer -> backbone2 -> ISP2

Really? 

Even the optimistic case:

ISP1 -> private backbone -> NAP -> ISP2

Could be a lose if the congestion at the NAP is bad enough.
 
>The idea that 'Real backbones" don't connect where the connecting is
>happening is pretty strange.  Real backbones connect to a lot of places.

Yes they do.  But they don't do *all* their peering at the NAP's.  A 
NAP peer is not as good as private peer.  It's better than no peer but 
if the there is enough traffic to justify it, it is better to use a 
private peer.  And bigger backbones are more likely to have enough cross
traffic to justify private peers among them.  A backbone provider that
does all of their peering at NAP's is at a severe disadvantage to larger
backbones who use private peering.

-- 
Real courtesy requires human effort and understanding.  
Never let your machine or your habit send courtesy copies.