[87301] in North American Network Operators' Group
Re: Two Tiered Internet
daemon@ATHENA.MIT.EDU (Per Heldal)
Wed Dec 14 06:58:54 2005
From: "Per Heldal" <heldal@eml.cc>
To: nanog@merit.edu
In-Reply-To: <OF68A1F92A.5303C2BD-ON802570D7.003B43B0-802570D7.003BECAA@btradianz.com>
Date: Wed, 14 Dec 2005 12:58:23 +0100
Errors-To: owner-nanog@merit.edu
On Wed, 14 Dec 2005 10:54:43 +0000, Michael.Dillon@btradianz.com said:
>
> But there is another way. If you provide enough bandwidth
> so that your peak traffic levels can travel through the
> network without ever being buffered at any of the core
> network interfaces, then everybody is a king. If you charge
> your customers a higher fee for such a network than your
> competitors do, then we have a tiered Internet. This
> unobstructed network was pioneered by Sprint on it's
> zero-CIR frame relay network and they carried this forward
> into their IP network as well. Other companies have
> carried forward this architecture as well.
That's the way all serious providers did IP-backbone engineering when
there was no QoS. Local congestion in the access-network would happen
from time to time even back in the 90s, but a network with
congestion-problems in the backbone would soon be a network with no
customers. Even today, it's the superior principle for backbone
engineering. Most QoS-handling (and other traffic-engineering) gizmos,
although some look good on paper, are too complex and too
labour-intensive to offer cost-saving or other operational advantage in
large IP backbones. Bandwith in the form of long-haul dark-fiber or
colors would have to be much more expensive to change that equation.
//per
--
Per Heldal
heldal@eml.cc
