Load sharing between two ISPs

[Learner09]

We have two edge routers from two ISPs with static default routes for internet, both the routers are connected to a layer 3 switch. Currently we have configured HSRP between two routers with default route in the switch towards the virtual IP to achieve active passive scenario between the two ISPs, now we have a requirement to make them as active active so that the traffic load will be shared between the two ISPs. How can we achieve this? Thanks in advance.

[MHM Cisco World]

Config hsrp group 

https://www.cisco.com/c/en/us/support/docs/ip/hot-standby-router-protocol-hsrp/13781-7.html

MHM

[balaji.bandi]

If you looking Dual Active/Active

There are 2 Options.

1. You can introduce IGP between Switch and Router

2. Add both the Subnet in NAT ACL so they can NAT the Traffic.

3. Track with SLA based on the Route Failure and shutdown the Router interface on Lan side / bring down the OSPF / or play with cost so other ISP which is UP take prority for all the traffic.

[Learner09]

@balaji.bandi @MHM Cisco World Thank you for your replies and suggestions, we have some complexity in splitting the LAN subnets and implementing MHSRP, so I have tried a lab setup with OSPF in the routers and switch. After injecting the default route into the ospf from the routers, I am able to reach internet from the switch and end devices, but they are always preferring the same ISP and not load sharing even though the ospf metrics/costs are the same. Kindly advice

[MHM Cisco World]

Are HSRP active router  connect to two ISP or one?

[Richard Burts]

We might be able to give better advice if we had more detail about what you have done in your lab setup. Could you post the configs? Also it would be helpful if you would post the output of show ip route from the switch.

[balaji.bandi · Accepted solution]

You need to post the show run configuration from switch. if that is equally Load-balance traffic should go both the sides, also as suggested you can also do Static route (but make sure you need to remove HSRP.)

[Joseph W. Doherty · Accepted solution]

From what you describe, possibly it might be as simple as L3 switch having two default routes, one toward each ISP.

[Joseph W. Doherty]

BTW, you could use MHSRP as targets for the two default routes, which might be a simple addition to what you've been doing.  @MHM Cisco World  initial reply's reference might explain further.

[paul driver]

Hello
Just like to add - Based on your OP it seems you may have various way to accomplish this task however without knowing the full extent of the topology in terms of the routing it could prove difficult to provide a definitive answer.

However, one possible solution may be.

Ingress traffic - advertise summary routes to each ISP and then more specific routes to each ISP for the preferred ingress traffic pertaining to each lan network

Egress traffic -  have default route pointing to the primary ISP  for ALL traffic but then policy base route towards the secondary ISP for your specific lan traffic to route via that secondary ISP

[Richard Burts]

Paul

When I read the original post I also thought it important to address Ingress traffic and egress traffic separately. Your suggestion for ingress traffic to advertise a summary to both and then more specific routes to each ISP requires that the customer have at least a /23 Public address block (you can not advertise to an ISP anything smaller than /24). And I doubt that they have that much Public IP block. So I am not sure that your suggestion is a good fit for this customer.

As I read the OP I get the impression that the current environment is based on static routing. So some suggestions might well be based on maintaining that environment (such as PBR to send some traffic to the second ISP). Other suggestions might be based on implementing dynamic routing (perhaps dynamic within the customer network and static to ISP, or perhaps dynamic within the customer network and BGP between customer and ISP). I believe it is much easier to arrange for redundancy and failover when using dynamic routing. If the OP will indicate which direction they would want to go then we can make better suggestions.

[paul driver]

Hello @Richard Burts 
nw mate it was only a possible suggestion obviously network dependent.

[Learner09]

Thank you all for a valuable inputs and suggestions, above is my lab setup and attached are the configurations. Below are the default routes from the switch, though the distance and metric are the same for both the paths, the trace to google DNS takes the same path no matter how many times I try. Here I have used two static default routes as suggested by @Joseph W. Doherty  above, and I have the same results when using OSPF as well.

Switch#sh ip route 0.0.0.0
Routing entry for 0.0.0.0/0, supernet
Known via "static", distance 1, metric 0, candidate default path
Routing Descriptor Blocks:
* 10.50.20.1
Route metric is 0, traffic share count is 1
10.50.10.1
Route metric is 0, traffic share count is 1

VPCS> trace 8.8.8.8
trace to 8.8.8.8, 8 hops max, press Ctrl+C to stop
1 10.50.50.1 2.921 ms 2.208 ms 3.167 ms
2 10.50.20.1 4.594 ms 3.306 ms 2.921 ms
3 192.168.8.1 7.760 ms 5.568 ms 4.755 ms
4 192.168.100.1 8.003 ms 10.890 ms 4.918 ms
6 10.188.193.64 9.785 ms 11.079 ms 13.131 ms
7 10.188.193.21 13.202 ms 13.381 ms 10.027 ms
8 10.188.6.202 13.151 ms 12.077 ms 10.974 ms

VPCS> trace 8.8.8.8
trace to 8.8.8.8, 8 hops max, press Ctrl+C to stop
1 10.50.50.1 2.084 ms 3.072 ms 1.646 ms
2 10.50.20.1 4.066 ms 3.035 ms 3.092 ms
3 192.168.8.1 6.525 ms 6.165 ms 5.729 ms
4 192.168.100.1 26.712 ms 6.344 ms 8.000 ms
6 10.188.193.76 10.336 ms 9.832 ms 12.312 ms
7 10.188.193.45 10.639 ms 9.808 ms 9.441 ms
8 10.188.6.202 11.188 ms 13.056 ms 11.464 ms

[MHM Cisco World · Accepted solution]

Route metric is 0, traffic share count is 1
that meaning the load per-destination
convert it to per packet  

add this under the interface toward ISP1 and ISP2 edge router 

Router(config-if)#no ip route-cache

then check again

[MHM Cisco World]

Or use cef load balance per packet with 

Ip cef load-sharing per-packet 

MHM