Equal Load Balancing between two E3 link

[arabindas]

Hello Experts,

In our organization, we are planning to have 2 E3 links to connect to one of our new site. At both of our site we have two core switches with HSRP configured and the routes are being advertised using EIGRP. So for better backup i was planning to have the 2nd link on the secondary switches at each site.

We want both the links to load balance the traffic between the two locations. So for that I was wondering if the EIGRP configured at both site core switches would take care of equal load balancing, without having to do any special configuration.

Please let me know if any additional configuration to have the equal traffic load balancing would work in this scenario. Attached is the design.

Thanks

Arabinda

[Giuseppe Larosa]

Hello Arabinda,

EIGRP is fine use GLBP instead of HSRP to achieve real outbound load balancing or only one link will be used (the one connected to the HSRP Active Router) in that direction site to central office

Hope to help

Giuseppe

[Joseph W. Doherty]

As Giuseppe describes, if you can use GLBP instead of HSRP, it should balance traffic across both links. You also have two other options.

First, you can tweak costing on your primary router(s) to make the alternate path have the same cost as the primary path. Then EIGRP will equal cost load balance.

Second, EIGRP supports unequal cost load balancing. With extra EIGRP configuration, it will use both links. (NB: others have posted on these forums that they believe unequal cost EIGRP routing can impact the device's CPU.)

Which is better?

The GLBP solution is likely the easiest to implement. (NB: changing from HSRP to GLBP changes the virtual MAC and I don't believe GLBP performs a gratuitous ARP. I.e. active hosts can be confused by the change.)

The EIGRP solution load balances (I believe) by active flows where GLBP balances by MAC, so your load distribution might be somewhat better.

[Giuseppe Larosa]

Hello Joseph,

for me playing with EIGRP is not an option if there are only client Vlans the client PCs send traffic to the default gateway.

With HSRP all of them send traffic to the same device.

Making that switch to see two equal cost paths via its connected E3 interface and via the colocated node for me it is difficult.

Without any change the route from the colocated node is not a feasible successor because it is not less then FD but it is AD = FD and so no load balance can happen.

Making the path equal from the point of view of the first node will make the other node the preferred path for inbound traffic defeating load balancing in the opposite direction from HQ to site (if possible)

Hope to help

Giuseppe

[Joseph W. Doherty]

Giuseppe, you may be correct about the tweaking load balancing for equal costing's impact to the other (reverse) direction. I know it would work for OSPF, but didn't realize EIGRP can't cost asymmetrically.

I also agree, tweaking costing in this fashion, even if it would work, is a nuisance (perhaps more so than difficult and is why I noted GLBP would be easier). But if it can be done, it should work. (I know it does for single subnets/VLANs with OSPF. Did it when necessary before there was GLBP. [Also, now, often just use GLBP for such situations, but that also assumes you have Cisco equipment that supports it. For example, 3560/3750s used for metro WAN links only support HSRP, I believe.])

You're correct about "feasible successor" (etc.) without any additional changes, EIGRP would not use the alternate path, but I did note additional configuration would be required.

Assuming unequal cost routing is properly configured, is there some reason you see it would not work starting from the same gateway router? The only issue, besides possible CPU impact, that comes to my mind, it should distribute the flows based on relative costs, and the secondary path would get less traffic. But also assuming LAN link between the two WAN routers, I also assume the cost delta would be very little and result almost in a 50/50 split (actually would be more variable due to actual flow loads).

[Giuseppe Larosa]

Hello Joseph,

I couldn't make a test but I reviewed EIGRP route TLV that contains the advertised distance = metric from the point of view of the router.

The tweaking of the metric should be successful if the delay of E3 link of R1 is incremented so that R2 AD for all HQ routes can satisfy the Feasibility condition.

Then using variance is possible to get usage of both E3 links.

The other direction shouldn't be affected as I was afraid: from the point of view of core routers they receive the route TLV from R1 and R2 that define the client vlan subnet and to calculate the metric from their point of view they use the delay as defined on their side of E3 interfaces.

It would be nice if someone could test this.

Thanks for this useful discussion

Hope to help

Giuseppe