BGP

[Siddique]

[MHM Cisco World]

Understand Load Share with BGP in Single and Multihomed Environments - Cisco

this link help you

[Kanan Huseynli]

Hi,

In real world, you should also consider NAT. If you have PA (provider aggregatable) address space, then you should do conditional (policy-based) NAT based on chosen egress interface (with route-map, match interface and address).

If you have PI (provider independent) scope, depending on scope range you may or may not do load sharing. For instance, if you have one /24 PI subnet, you normally advertise it via primary ISP (normal advertisement) and secondary ISP (with as-path prepend). Then traffic should leave and return primary ISP to avoid asymmetric routing with respect to connected ISPs (they enable DPI most of the time, however you can ask to disable also).

If you have scope more than /24 then you can divide into (at least) two parts (for example two /24 from one /23) and advertise the first scope to ISP1 (normal), but with prepend to ISP2 (the second scope will be vice versa).

[Richard Burts]

The original post tells us what result is desired but tells us nothing about the environment or about how the router is configured. The diagram suggests that the router is running BGP with both ISP. Does each ISP advertise a default route? Does either ISP advertise more than just a default route. Does the router config do any manipulation or filtering on routes received from either ISP? Does the router do any manipulation or filtering on routes advertised to either ISP?

Assuming that it is a simple configuration of BGP (no manipulation or filtering of routes) and that both ISP advertise a default route then configuring BGP with neighbor statements for both ISP should achieve the objective. I agree with previous post that it is also required to configure NAT for outgoing traffic, and probably NAT with route map which can match the NAT with the outgoing interface.

[Siddique]

Received only default route from both ISP. 

Advertising only 1.1.1.0/ 23 towards both ISP.  

No any manipulation configure yet.  

[Kanan Huseynli]

Hi @Siddique ,

so, normally you do advertise /23 to both ISP. In this case outbound traffic will be based on load-balancing algorithm running on device and it either choose ISP1 or ISP2 (assuming you have both default route in routing table). Return traffic, can not be determined, it will be either ISP1 or ISP2, depending on remote location routing decision or connected ISP routing decision.

Sometimes, it results asymmetric routing in internet that means traffic leaves ISP1 but returns ISP2. If DPI or another inspection enabled on ISP network, you will have problem in that traffic flow.

I'd suggest to split /23 into two /24 and advertise the first segment to ISP1 and ISP2 (but with as-path prepend). The second segment to ISP2 and ISP1 (but with as-path prepend). You should additionally advertise /23 to both without any manipulation. /23 will be used if one of the ISP fails, normally /24 wins due to longest-match.

In this case, if traffic leaves ISP1, it will be NATed to the first /24 segment (you should do NAT of course), then return traffic will be via ISP1 (because you do /24 on the second ISP and /24 is longest match than /23 which is advertised to both). Similarly, happens to the second /24 subnet.

[MHM Cisco World]

receive only default route and you have only one edge router then 
use weight to prefer the default route from one ISP, and if it failed then the router will shift to other ISP. 
for NAT not need anything the NAT (overload) will use the egress interface you use.
for static NAT you only need conditional NAT 
for weight 
neighbor <ISP1> route-map default in <<- ISP1 will be primary and ISP2 will be backup
route-map default permit 10
set weight 50000
 

[Joseph W. Doherty]

So, your router shows both ISPs as next hop for default route?

If not, if your default route is via BGP, you may need to configure BGP to use multiple path and Cisco's hidden relax command.

If so, by default, your router should, more or less, round robin flows outbound.  I.e., you should obtain, over time, about a 50/50 load share.

BTW, understand, although your outbound load sharing can send flows on the less optimal path, although "optimal" using BGP is debatable.

For inbound, though, and as mentioned by others, path usage will depend on what sender believes is best path to you.

There are "techniques" that can be used to try to influence or control inbound path selection, but when working with the Internet, IMO, add bandwidth to the heavier used inbound path.

To your fail over question, BGP, alone, should be able to deal with that.