[c-nsp] BGP transit selection from source customer network

P C pc50000 at gmail.com
Wed Aug 3 13:37:26 EDT 2011


Only PBR is really required to make a certain "netblock/source"
address go to a particular egress interface.  It should work as
desired

But be warned... PBR/"source based routing" is process switched on
some platforms and results in a large performance hit.  It can also
suffer from some redundancy problems if not properly implemented (if
ISPB goes down, how will PBR know to re-route?)

It's also way too "static/inflexible" for my liking since it's not
destination based -- when equipment is changed/upgraded down the road
-- or even an interface moved -- it's easy to "miss" that one
source-based routing entry and cause an outage for that customer

On Wed, Aug 3, 2011 at 6:40 AM, Alexandre Durand
<alexandre.durand at tasfrance.com> wrote:
> Well I don t agree, It is easy enought on route A to overweight ISPA so path
> is preffered over ISPA for any source, so I just use PBR to send my customer
> traffic via ISPA and the other source traffic to the Router B with ISPB. As
> this is said, I might change my route-map as follow:
>
> ip prefix-list cust1 deny x.x.x.x/24
> ip prefix-list cust1 permit 0.0.0.0/0 le 32
>
> route-map cust1, permit, sequence 10
>  Match clauses:
>   ip address prefix-list cust1
>  Set clauses:
>   ip next-hop ISPB
>  Policy routing matches: 15 packets, 1710 bytes
>
> Any traffic except my customer traffic will be forwarded to ISPB
>
>
>
>
> On 03/08/11 09:54, Andrii Morozov wrote:
>>
>> According to the scheme drawn above, you have an iBGP between your border
>> routers. Thus, they will send the best routes to each other. This means, no
>> matter whether you set PBR or not, if the best path is selected over ISB B,
>> the traffic will traverse that link, but not ISP A.
>>
>> 2011/8/2 Alexandre Durand <alexandre.durand at tasfrance.com
>> <mailto:alexandre.durand at tasfrance.com>>
>>
>>    weird I used only PBR and it seems to be working. I created a
>>    route-map that send cust1 traffic out to ISPA (RIB internet
>>    routes) and any other customer source traffic out to ISPB. the
>>    weird thing is that ISPB routes are NLRI routes and not RIB routes
>>    ... how can this be working ...
>>
>>    regards
>>
>>    alexandre
>>
>>    route-map cust1, permit, sequence 10
>>     Match clauses:
>>       ip address (access-lists): cust1
>>     Set clauses:
>>       ip next-hop ISPA
>>     Policy routing matches: 122 packets, 13908 bytes
>>    route-map cust1, permit, sequence 20
>>     Match clauses:  (ANY)
>>     Set clauses:
>>       ip next-hop ISPB
>>     Policy routing matches: 15 packets, 1710 bytes
>>
>>
>>    On 02/08/11 11:04, Alexandre Durand wrote:
>>
>>        Hi,
>>
>>        You are right, however on router A is connected to ISPA, bgp
>>        will add only preferred ISPA routes and  not the others routes
>>        from router B ISPB with IBGP, but only ISPA routes will be
>>        preferred on this router A (attribute weight), router A will
>>        be not aware about ISPB rouites unless ISPA goes down.
>>
>>        I may use a dedicated vrf for ISPA so global routing table and
>>        vrf table are not bound each other. And then use PBR over
>>        route-map to redirect first via ISPA VRF and secondly via
>>        global routing table where ISPB routes will be present in RIB.
>>        What do you think about it?
>>
>>        The only issue witch such solution is that I will need to
>>        upgrade the ios because vrf under route-map is not supported
>>        on my IOS(12.2(18)SXD3).
>>
>>        regards
>>
>>        Alexandre
>>
>>        On 01/08/11 17:56, David Freedman wrote:
>>
>>            -----BEGIN PGP SIGNED MESSAGE-----
>>            Hash: SHA1
>>
>>            On 01/08/11 15:35, Alexandre Durand wrote:
>>
>>                Hi Andrew,
>>
>>                Tnhan you for you answer. Actually youa re right about
>>                the inbound
>>                traffic in which I can easily advetise this network
>>                with a preference to
>>                my "ISPA", this is fine and already done.
>>
>>                The issue here is about the oubound traffic, when my
>>                customer traffic is
>>                going back out to ISP, I can also you weight on the
>>                router where I host
>>                this "ISPA" and use no-advertise attribute so I don t
>>                advertise
>>                prefixes to others iBGP routers, fine but others
>>                customer traffic will
>>                also be routed via ISPA because the shortest ISP path
>>                is going throught
>>                this router too...
>>
>>            I think what you are saying sounds like this:
>>
>>            |Your Network---Internet---------|
>>
>>            [Cust1]--( N )
>>                     ( E )--[ISPA]-(N)-[Cust4]
>>            [Cust2]--( T )         (E)
>>                     ( W )--[ISPB]-(T)-[Cust5]
>>            [Cust3]--( K )
>>
>>
>>
>>            I take it you want to do :
>>
>>            - - Predetermined outbound path Cust1->Cust4 via ISPA
>>            - - Predetermined outbound path Cust2->Cust5 via ISPB
>>            - - Cust3 ->  Cust4/5 via any/best path
>>
>>            What you are talking about is essentially policy routing
>>            at your edge
>>            (i.e using your own criteria, like customer ID to
>>            influence outbound
>>            routing, not metrics)
>>
>>            you can do this simply with PBR at the edge but it won't
>>            be very
>>            reliable, but you might want to look at OER/PFR
>>
>>
>>  (http://www.cisco.com/en/US/products/ps8787/products_ios_protocol_option_home.html)
>>
>>
>>            Dave.
>>            -----BEGIN PGP SIGNATURE-----
>>            Version: GnuPG v1.4.10 (GNU/Linux)
>>            Comment: Using GnuPG with Mozilla -
>>            http://enigmail.mozdev.org/
>>
>>
>>  iEYEARECAAYFAk42zL4ACgkQtFWeqpgEZrK2BgCgwGU7nRzRPklKNII9L69uA00I
>>            w5YAoKkhzi8L8Roo5oONsg/6z1aNW0ka
>>            =lLQt
>>            -----END PGP SIGNATURE-----
>>            _______________________________________________
>>            cisco-nsp mailing list cisco-nsp at puck.nether.net
>>            <mailto:cisco-nsp at puck.nether.net>
>>            https://puck.nether.net/mailman/listinfo/cisco-nsp
>>            archive at http://puck.nether.net/pipermail/cisco-nsp/
>>
>>
>>
>>
>>
>>    --     Alexandre DURAND
>>    TAS FRANCE
>>    WTC 1-K, 1300 route des Crêtes
>>    06560 Valbonne Sophia Antipolis
>>    Phone :    +33 (0)4 92 94 56 93
>>    Fax   :    +33 (0)4 92 94 33 99
>>    Web: http://www.tasfrance.com
>>    Email : alexandre.durand at tasfrance.com
>>    <mailto:alexandre.durand at tasfrance.com>
>>    peering: http://as8554.peeringdb.com
>>
>>    _______________________________________________
>>    cisco-nsp mailing list cisco-nsp at puck.nether.net
>>    <mailto:cisco-nsp at puck.nether.net>
>>    https://puck.nether.net/mailman/listinfo/cisco-nsp
>>    archive at http://puck.nether.net/pipermail/cisco-nsp/
>>
>>
>
>
> --
> Alexandre DURAND
> TAS FRANCE
> WTC 1-K, 1300 route des Crêtes
> 06560 Valbonne Sophia Antipolis
> Phone :    +33 (0)4 92 94 56 93
> Fax   :    +33 (0)4 92 94 33 99
> Web:       http://www.tasfrance.com
> Email :    alexandre.durand at tasfrance.com
> peering:   http://as8554.peeringdb.com
>
> _______________________________________________
> cisco-nsp mailing list  cisco-nsp at puck.nether.net
> https://puck.nether.net/mailman/listinfo/cisco-nsp
> archive at http://puck.nether.net/pipermail/cisco-nsp/
>



More information about the cisco-nsp mailing list