Localizing traffic using BGP

Fergie (Paul Ferguson) fergdawg at netzero.net
Thu Mar 3 04:25:59 UTC 2005

Well, the most specific prefix wins in the forwarding
selection process, but I won't make any comments on
"best current practice" since "best" is somewhat
relative to the idea of max-aggregation ....but
this reference might help:

G. Huston
Request for Comments: 3221
Commentary on Inter-Domain Routing in the Internet


5.4  Lack of Common Operational Practices

   There is considerable evidence of a lack of uniformity of operational
   practices within the inter-domain routing space.  This includes the
   use and setting of prefix filters, the use and setting of route
   damping parameters and level of verification undertaken on BGP
   advertisements by both the advertiser and the recipient.  There is
   some extent of 'noise' in the routing table where advertisements
   appear to be propagated well beyond their intended domain of
   applicability, and also where withdrawals and advertisements are not
   being adequately damped close to the origin of the route flap.  This
   diversity of operating practices also extends to policies of
   accepting advertisements that are more specific advertisements of
   existing provider blocks.


- ferg

-- Ashe Canvar <acanvar at gmail.com> wrote:

Hello all,

BGP design question. My scenario is very much like an online gaming
company with servers on multiple continents.

Details :
 1. I have datacenters/servers in many POPs globally. 
 2. I am a stub AS with  multiple ISPs available at each POP. 
 3. I do not have the resources to buy dedicated fiber to link these
POPs, so my backbone is essentially a bunch of VPN tunnels.
 4. I own a /21. Of which I announce /24's.
 5. I have a mechanism to allocate client sessions to any ONE of the
POPs based on network performance (which is what I want to optimize).

Questions :

 I presently announce different /24's out of different POPs to
localize traffic. All these prefixes are announced from the same AS
number. This obviously can't be best practice. Are there any published
BCPs or RFCs tackling this issue ?

 Is it recommened that I get multiple AS numbers and announce
symmetrically from each "smaller" AS ? Or should I announce
inconsistently from different locations ?


"Fergie", a.k.a. Paul Ferguson
 Engineering Architecture for the Internet
 fergdawg at netzero.net or
 fergdawg at sbcglobal.net

More information about the NANOG mailing list