mpls primary + backup, two ISPs: considerations

[ariela]

Hi folks,

I need your advice ;)

I've an hub-2spokes topology, and the Customer has to implement a primary link MPLS with a provider, and a backup link MPLS with another ISP. All sites have an isolate routing process ospf (area 0) and two CE, one for each link.

So, depending that, at the moment, a load balancing solution is not a 'must', but maybe a 'nice to have', I'm thinking about these scenarios:

a) CE-PE iBGP

b) CE-PE OSPF (superbackbone for primary link; for the backup link I could use a traditional OSPF-BGP redistribution, so we have LSA3 from primary, and LSA5 from backup)

Any advice will be appreciated

Regards

Andrea

[Joseph W. Doherty]

What we've done is:

CE-PE EBGP

CE to local LAN, OSPF

CE-CE IBGP (private AS) and OSPF

We redistribute BGP into OSPF, we provide site summary address to EBGP via network statement and static route to null.

We control LAN to CE traffic by OSPF costing. Normally, equal cost, so we load balance. If no internal LAN routers, CEs run GLBP.

Each CE knows about other ISP routes, but prefers its CE-PE link. If link goes down, traffic will flow to other CE.

For most sites, we don't leak one ISP routes to the other.

[ariela]

Hi,

thanks for your answer. Someone could discuss with me the two scenarios in attach (pros, cos, feasibility)?

Thanks again for your support

Regards

Andrea

[swaroop.potdar]

Andrea, looking at your ppt, and the topology

its evident that the major con to your IBGP/EBGP method is the administrative overhead (comapred to OSPF). But apart from that its much better an option than OSPF, as it gives you more control whether to load balance or have a primary backup. In case of OSPF its pure primary backup.

Also OSPF beats your orginal purpose of having primary backup plus load sharing if possible.

Although you can consider running OSPF with same area (area 0) at PE-CE with same domain id. and set the cost equal for all incoming routes. SO you can load balance as well as have backup.

Though do bear in mind that dont create a transit for either SP VPN through your site.

Apart form that there isnt much to weigh the pros and cons, as you can safely deploy either of your methods as they are commonly deployed.

HTH-Cheers,

Swaroop

[ariela]

hi Swaroop,

thanks for your answer.

"Though do bear in mind that dont create a transit for either SP VPN through your site."

Could you explain it better?

In case of OSPF Superbackbone, and dual-homed connection, there's the Down Bit that could prevent loops. In case of iBGP/eBGP, I've to prevent that a site could be a transit AS, correct ... what are you talkin about?

thanks again for your support

Regards

Andrea

[swaroop.potdar · Accepted solution]

Andrea, yes thats correct its not applicable for OSPF in this scenario where the down bit would take care of not advertising that route back. Its was more of a generic statement for any protocol used, to avoid suboptimal routing.

HTH-Cheers,

Swaroop