Service Providers behaviour for dual homed enterprises

Blake Hudson blake at ispn.net
Thu Sep 24 14:05:46 UTC 2015



Stephen Satchell wrote on 9/24/2015 8:39 AM:
> On 09/23/2015 02:38 PM, Jason Bullen wrote:
>> I've always worked in enterprise only so I thought you guys might be 
>> able
>> to help me with this one.
>> We are dual homed to Verizon and AT&T.  We prepend all our prefixes out
>> AT&T to make them least preferred.  During a recent issue we found some
>> users were coming in via AT&T.  Using various looking glasses it 
>> looks like
>> if I use an AT&T server(route-server.ip.att.net) the best path is the
>> prepended route through AT&T; in fact,I don't even see the VZB 
>> route.  If I
>> use a 3rd party looking glass(router-server.he.net) I see what I
>> anticipated, which is the shorter AS-Path through VZB.
>>
>> So if my research is correct, the internet prefers Verizon UNLESS 
>> they are
>> a direct AT&T customer then they would use the AT&T circuit.
>> Is this a standard practice that I should assume to encounter?
>>
>> Thanks in advance
>>
>
> That's been my experience, and with other sets of providers, too.
>
> My current company is dual-homed with AT&T and Charter Fiber. Those 
> customers on UVerse come in the AT&T link no matter what we do with 
> BGP to convince the cloud to let packets come in the fatter pipe.

Jason, while others have offered acknowledgement of the behavior you are 
seeing as well as solutions, I think it might be relevant to point out 
that this is simply a matter of BGP best path selection. BGP does not 
use AS path length (hops) as its primary path selector. Search for "bgp 
best path selection" to find out more about how BGP selects the best 
path. As others have noted, local pref is often utilized to control 
routing and should be your preferred way to control path selection in 
addition to AS path length. However, the ultimate way to control routing 
would be to advertise more specific prefixes via the path that you want 
traffic to flow.

--Blake



More information about the NANOG mailing list