From: R. Benjamin Kessler (bk-lists@kesslerconsulting.com)
Date: Tue Oct 08 2002 - 18:35:00 GMT-3
Thanks Howard; I'm going to enter "tweak-mode" and see what happens?
Provider 3 is the "better connected" of the three; its over a burstable
DS-3
The other two generally have an AS-path that is longer but they have a
"larger pipe." One is a 100Mb Ethernet non rate-limited, the other is a
Fast-E throttled down to 10Mb.
I'm just trying to "optimize" the connectivity as best we can; probably
more of an academic exercise than anything else.
Thanks for the input.
Ben
-----Original Message-----
From: nobody@groupstudy.com [mailto:nobody@groupstudy.com] On Behalf Of
Howard C. Berkowitz
Sent: Thursday, October 03, 2002 7:22 PM
To: ccielab@groupstudy.com
Subject: RE: Internet BGP Configuration "best practices"
>No Takers, trying again....
There's no really simple answer. I'm not being sarcastic when I say I
had to write a book dealing with a lot of the analysis involved, and
it doesn't cover everything. In this case, I'd need to know a lot
more to try for an optimized solution, including speed and delay of
the ISP links, connectivity of the ISPs, etc.
Communities are good and best current practice for most things like
this.
A good free reference is Avi Freedman's NANOG presentation at
http://www.nanog.org/mtg-9901/ppt/bgp102/index.htm
I have a related presentation at
http://www.nanog.org/mtg-0102/exterior.html
>
>-----Original Message-----
>From: nobody@groupstudy.com [mailto:nobody@groupstudy.com] On Behalf Of
>R. Benjamin Kessler
>Sent: Wednesday, October 02, 2002 9:01 AM
>To: ccielab@groupstudy.com
>Subject: OT: Internet BGP Configuration "best practices"
>
>Sorry for the OT; I'm looking for feedback on what people are doing in
>the real-world regarding BGP "load-sharing" for multiple Internet
links.
>
>Here's the scenario (obviously the addressing has been changed to
>protect the guilty :-)
>
>+--------------+ +--------------+ +--------------+
>| Provider 1 | | Provider 2 | | Provider 3 |
>| AS 65501 | | AS 65502 | | AS 65503 |
>+--------------+ +--------------+ +--------------+
> | | |
> | | |
> | | |
> +---+------------------+------------------+--------+
> | | | | |
> | \ / | |
> | \ / | |
> | \ / | |
> | \ / | |
> | +--------------+ +----------+ |
> | | Router 1 |---------------| Router 2 | |
> | +--------------+ +----------+ |
> | AS 65499 |
> +--------------------------------------------------+
>
>We (AS65499) are advertising one network (a /24 - for example say
>192.168.100.0/24) and accepting full routes from each of the providers.
>We're a non-transit AS!
>
>The link to 65501 is larger than that to 65502; the link to 65503 is
the
>smallest but that provider is generally "better" connected to sites on
>the Internet (i.e. when comparing the length of AS hops)
>
>Currently, I'm using AS-Path ACLs to set the local pref on routes
>received from each provider such that any routes originated in their AS
>or those one AS-hop away will have their local preference increased.
>Here's an example:
>
>ip as-path access-list 101 permit ^65501_[0-9]*$
>ip as-path access-list 102 permit ^65502_[0-9]*$
>ip as-path access-list 103 permit ^65503_[0-9]*$
>
>The following route-maps are applied as appropriate:
>
>route-map AS65501-IN permit 10
> match as-path 101
> set local-preference 150
>!
>route-map AS65501-IN permit 20
>
>route-map AS65502-IN permit 10
> match as-path 102
> set local-preference 140
>!
>route-map AS65502-IN permit 20
>
>route-map AS65503-IN permit 10
> match as-path 103
> set local-preference 130
>!
>route-map AS65503-IN permit 20
>
>I know that one of the providers tags each of its customer routes with
a
>community so I may change that route-map to match on that instead of
the
>as-path acl.
>
>Additionally, I'm going to advertise a metric out each link to try and
>influence the reverse path (since in-bound traffic is far greater than
>outbound).
>
>We've done prepends in the past to try and influence inbound traffic
but
>was hoping for a more elegant approach.
>
>I'd be interested in hearing what has worked for others on the list.
>
>Thanks,
>
>Ben
This archive was generated by hypermail 2.1.4 : Tue Nov 05 2002 - 08:35:42 GMT-3