[142159] in North American Network Operators' Group
SLA covering 3rd party assets, financial incentives
daemon@ATHENA.MIT.EDU (Patrick W. Gilmore)
Sun Jun 19 18:51:37 2011
From: "Patrick W. Gilmore" <patrick@ianai.net>
In-Reply-To: <114351.1308520030@turing-police.cc.vt.edu>
Date: Sun, 19 Jun 2011 18:50:50 -0400
To: NANOG list <nanog@nanog.org>
Errors-To: nanog-bounces+nanog.discuss=bloom-picayune.mit.edu@nanog.org
On Jun 19, 2011, at 5:47 PM, Valdis.Kletnieks@vt.edu wrote:
> On Sun, 19 Jun 2011 03:15:09 CDT, Robert Bonomi said:
>=20
>> Anybody got draft language for a SLA clause that requires routing 'at =
least
>> one hop _past_ the provider's network edge' for every AS visible at =
major
>> public peering points and/or LookingGlass sites? =20
>=20
> *every* ASN? Oh my. ;)
Many people are interested in things like end-to-end performance, or =
global connectivity. However, one must be careful how such an SLA is =
written. No network wants to guarantee performance or even simple =
connectivity to gear they do not own / control. (Well, almost no one. =
I know at least one company that does, but they don't sell "transit" =
per-se.)
Put another way, how do you write this such that my competitor cannot =
cause me to go bankrupt with SLA credits?
Some simple things spring to mind, such as: "Do substantially all =
prefixes appear in the table handed to BGP customers?" (Lawyers can =
fight over "substantially all". :) But is that really enough? Having a =
prefix in the table means nothing, if the path is over a cable modem in =
Sri Lanka. And the reverse, my prefixes appearing in other networks' =
tables, is under the control of my competitors.
I'm not saying it is impossible. I'm saying be careful.
Likely economic pressure is more productive, i.e. vote with your wallet. =
Unfortunately, on the Internet, we have a history of doing the =
opposite. When Sprint literally disconnected from some parts of the =
'Net with ACL 112, people intentionally bought from Sprint to ensure =
they could reach the entire Internet. When InternetMCI couldn't connect =
to an exchange without packet loss to save its life, people =
intentionally bought from InternetMCI to avoid the congestion. Etc., =
etc. What did we think these networks would do when we literally paid =
them for their faults?
Worse, if there is a network who will not peer, and a network who will, =
most people buy from the non-peering network. The result of this is =
more networks want to close down peering, fewer want to open it up. =
Seems counter-productive to me, and trivially easy to fix. For =
instance, make peering a requirement of transit purchases. Of course, =
there would be other requirements (still need to run a good network, =
24/7 NOC, fair pricing, yadda, yadda), but it uses financial incentives =
to promote the activities we want, not the opposite.
Perhaps it is time we stopped enabling - rewarding! - the very networks =
& behaviors most of want to change?
--=20
TTFN,
patrick
