[131592] in North American Network Operators' Group
Topic: Inter-AS BGP Local Preference Matrix
daemon@ATHENA.MIT.EDU (Rettke, Brian)
Fri Oct 29 12:55:15 2010
From: "Rettke, Brian" <Brian.Rettke@cableone.biz>
To: "nanog@nanog.org" <nanog@nanog.org>
Date: Fri, 29 Oct 2010 09:55:06 -0700
In-Reply-To: <mailman.1.1288353602.18241.nanog@nanog.org>
Errors-To: nanog-bounces+nanog.discuss=bloom-picayune.mit.edu@nanog.org
My company is building a national backbone network, leveraging leased lines=
and dark fiber from Tier 1/2/3 providers. What we've found is that when we=
buy IP in the major markets, our traffic does not flow deterministically w=
ith AS-Path as the metric. This is because most of the major providers give=
their customers one local preference value, and their peers another, in an=
effort to ensure SLAs are met by keeping customer traffic on-net for as lo=
ng as possible. There are varying values assigned, and some vendors don't o=
ffer community values to neutralize this effect.
I'm wondering if anyone has dealt with this in the past, or if it would be =
possible to have some sort of agreement on local preference manipulation. S=
omething similar to the below:
1. All vendors must offer at least 5 community values for local preference.=
This is to allow for customer-based multivendor traffic engineering.
2. All vendors must offer a local preference community value greater than t=
heir best default metric.
3. All vendors must offer a local preference community value lesser than th=
eir worst default metric.
4. All vendors should offer a range of community values both above and belo=
w local preference 100.
5. All vendors should make an effort to standardize the values/value ranges=
offered with other vendors.
6. All vendors should offer a local preference matrix to their customers, l=
isting the changes made to a specific AS (e.g. another vendor) to aid the c=
ustomer in making an intelligent routing decision for load balancing and tr=
affic engineering in a multivendor BGP environment.
It's obviously something that each of us would need to do individually, but=
I'm wondering if there is any way this could become a de facto standard, o=
r could be a method that the community at large could enforce somehow.
Sincerely,
Brian A . Rettke
RHCT, CCDP, CCNP, CCIP
Network Engineer, CableONE Internet Services
