One way to do this is to have both client fw/routers running in their 
own right, i.e. no carp failover.

Each router peers with one of the ISP routers via eBGP and then peers 
with it's partner via iBGP.

On each router use the 'weight' option to make each router believe it's 
learned routes are the best.

Each router will now install it's best route in the kernel routing table 
and believing it has the best route will also redistribute it's routes 
to the iBGP partner.

The result each router will have two routes to any network in it's BGP 
table, one via its eBGP which it regards as 'best' and another with a 
higher weight via it's partner router.

It's also important to tune the BGP dead timers as low as you can so 
that if a link is lost to an upstream BGP session is cleared as soon as 
possible minimizing the amount of black holed traffic. Once the BGP 
session is down the alternate route learned from the partner router will 
be used to replace the failed route in the actual routing table.

To control which route is used for outbound traffic CARP can be setup on 
the 'internal' interfaces. Which ever router is the master will be used 
as the egress point for the network. Padding the announcement to the 
secondary provider could also help with controlling incoming traffic, 
although in my experience the results are mixed.

Now I've never tried it on OpenBGP but on Cisco this works like a charm.

e.g.

[ISP1]            [ISP2]
  |                 |
 ebgp              ebgp
  |                 |
[PRIV1]---iBGP---[PRIV2]
  |                 |
  M                 S
  |                 |
  ---------|---------

All traffic would flow out of PRIV1 / ISP1, if PRIV1 or ISP1 failed 
traffic would flow out of PRIV2 / ISP2.





BARDOU Pierre wrote:
quoted text
>  
> Hello,
>
> Failover already works with BGP on my test conf, the problem is that BGP 
> only selects ONE route to a destination, so there is no load balancing.
>
> The easiest for me would be to tell BGP to keep TWO routes to each
> Destination, and use them in a round-robin way.
>
> That's what Cisco does with BGP multipath
> http://www.cisco.com/en/US/tech/tk365/technologies_tech_note09186a0080094431
> .shtml#bgpmpath
>
> But AFAIK there is no way to setup this with openBGP.
>
> Am I right ?
>
> --
> Cordialement,
> Pierre BARDOU
>
> -----Message d'origine-----
> De : cnav@talamasca.pl [mailto:cnav@talamasca.pl] 
> Envoyi : mercredi 8 octobre 2008 09:05
> @ : BARDOU Pierre
> Cc : Frans Haarman; misc@openbsd.org
> Objet : Re: OpenBGP load balancing between 2 ISP (multihoming)
>
> BARDOU Pierre wrote:
>   
>> Hello,
>>  
>> I can load balance on the firewalls with pf , but the problem of that 
>> Solution is that there is no failover AFAIK.
>> If I loose a link between an ISP and me half of the packets will be lost.
>>
>> And not loosing packets is more important to me than load balancing...
>>  
>> --
>> Cordialement,
>> Pierre BARDOU
>>  
>>
>> ________________________________
>>
>> De : Frans Haarman [mailto:franshaarman@gmail.com] Envoyi : mardi 7 
>> octobre 2008 18:54 @ : BARDOU Pierre Cc : misc@openbsd.org Objet : Re: 
>> OpenBGP load balancing between 2 ISP (multihoming)
>>
>>
>> 2008/10/7 BARDOU Pierre <bardou.p@mipih.fr>
>>
>>
>> 	Hello,
>> 	
>> 	I am trying to set up a configuraion like this :
>> 	
>> 	            +------- -+       +---------+
>> 	            |  ISP1   |       |  ISP2   |     Cisco
>> 	            | ROUTER  |       | ROUTER  |
>> 	            | AS3215  |       | AS12670 |
>> 	            +---------+       +---------+
>> 	                 |                |
>> 	                 |                |
>> 	            +---------+       +---------+
>> 	            |   BGP   |       |   BGP   |
>> 	            | ROUTER  |       | ROUTER  |     OpenBSD 4.3
>> 	            | AS47818 |       | AS45818 |
>> 	            +---------+       +---------+
>> 	                 |                |
>> 	                 |                |
>> 	            +-------------------------+
>> 	            |    217.109.108.240/28   |
>> 	            +-------------------------+
>> 	                 |                |
>> 	                 |                |
>> 	            +--------+        +-------+
>> 	            |   FW   |--------|  FW   |       OpenBSD 4.3
>> 	            | MASTER | pfsync | SLAVE |
>> 	            +--------+        +-------+
>> 	                 |                |
>> 	                 |                |
>> 	            +-------------------------+
>> 	            |     PRIVATE NETWORKS    |
>> 	            +-------------------------+
>> 	
>> 	I'd like to load balance outgoing connections to the internet,
>> 	but I don't know how to configure openBGPd to do this.
>> 	I searched a lot on the Internet and I found a lot of informations
>> 	on how to do this with cisco, but I have never found an openBGP
>> solution.
>> 	Some people speak about it but I have never seen it.
>> 	
>> 	I made a test conf where failover works like a charm (using iBGP on
>> the
>> 	FW's with 'set nexhop self' on BGP routers), but when both
>> connections
>> 	are active only one is used.
>> 	
>> 	Would it be possible to help me please ?
>> 	Is setting up iBGP sessions between FW's and BGP routers a good idea
>> ?
>> 	Should I rather use OSPF for this ?
>> 	And in tha case how to configure it to loadbalance/failover ?
>> 	
>> 	Many thanks
>> 	
>> 	PS : loadbalancing incoming connections too would be very nice, but
>> I
>> 	understood it was much more difficult.
>> 	
>> 	--
>> 	Cordialement,
>> 	Pierre BARDOU
>> 	
>>
>>
>>
>> just wondering......
>>
>> What happens when you load balance your
>> traffic on your firewalls ? So you devide
>> the traffic over both bgp routers:
>>
>> http://www.openbsd.org/faq/pf/pools.html
>>
>> maybe you could even do the route-to 
>> on the bgp routers ?
>>
>> something like:
>>
>> route-to { ($ext_if $ext_ISP1), ($local_if $BGP2 ) } round-robin 
>> from $lan_net to any keep state 
>> #and on the other bgp router 
>> route-to { ($ext_if $ext_ISP2), ($local_if $BGP1 ) } round-robin 
>> from $lan_net to any keep state 
>>
>> Beware: I have no idea if any of this is possible.
>> But thats what I'd try :)
>>
>> Gr. FH
>>
>>     
>
> If you want to use fail-over capability of bgp, you can use prepend to 
> increase length of one path. I have no experience with configuring 
> openbgpd but on juniper/cisco it seems to work great.
>
> Regards,
>   Marusz