SD Wan and MPLS load balancing

Anwar Safian · ‎04-08-2019

Hi expert

1) I would like to know why a lot of enterprise use MPLS active-standby mode? Why not active-active? Is it highly inefficient as enterprises are paying for far more bandwidth than they are actually leveraging.

2) What is the difference between MPLS and SD Wan load balancing? Which one is better?

Simon Ko · ‎05-16-2019

I hope I know the answer but here is what I think.

Usually, dual MPLS has a vrrp configured on two routers.

One router will become a master, and it will route all the traffics out of network.

In order to be a symmetrical routing, return packets will take the same path.

Second router is standby by design.

On SDWAN - Viptela, routing is not done on per interface basis, but on per site basis.

Meaning, to get to 10.20.30.0/24, next hop is a site, not ip address of next hop.

This means, multiple paths can all be used, based on policy.

If you have two routers in sdwan/viptela, primary router will be used for traffic out of network.

Return packets are usually routed through the second device.

In my case, I have two vEdge out of DC.

One with better MED setting will be used for all outgoing traffics.

Return packets are via the second vEdge device.

tzarski · ‎05-24-2019

Hi Anwar,

1. In dual-edge branch scenario we are using active/standby setup for features which requires symmetric traffic flow. For example most DPI engines require full flow information in order to properly identify applications based on signatures. This means that traffic needs to exit and enter through same physical box (unless you use advanced detection mechanisms like Cisco SD-AVC).

2. The SD-WAN load balancing methods are much more flexible and granular. Except standard per-flow load balancing we can achieve things like:

weighted unequal per-flow load balancing
per application traffic pinning: based on DPI app recognition, define critical apps to use MPLS, and rest to use INET transport
Application-Aware Routing: define per application load balancing based on SLAs (e.g run Webex on transport which has latency less than 100ms, or send O365 traffic through link with loss less than 4%)