BGP outbound load balance. Is the local preference the solution?

[michaelca99]

Background:

Our backbone has 100Mbps max capacity. Some links only can handle 50Mbps. The max load of our network could be around 120Mbps. Currently, the link1 is running and all traffic is passing through link1. We are using BGP with upstream ISP. We are going to setup second link to do load balance.

I knew I can use BGP local preference to control outbound traffic.

My questions:

Is the BGP local preference good solution in our scenario? Or Is there another better solution?

[jwbensley]

I don't know enough about BGP to suggest how you could do this via BGP. One way though to have to forwarding gateways is to use GLBP. It would be worth adding a link between SW1 and SW3 for this so this continue to function in the loss of SW2.

Another idea is (if you have a spare interace on each router) you can realitively cheaply add a link between your two routers. Add a BVI between up upstream link from PE1 into R1 and the link to R2 on a VLAN sup-interface. Do the same on R2 to R1 on a seperate VLAN. This way you can bring up a BGP peering to both upstream devices on both devices. If you set up HSRP between R1 and R2, R1 can forward traffic other both links using ECMP with HSRP to fail over to R2 if it fails, although you would lose one upstream device if either R1 or R2 fail.

If this topology you can achieve the same as above but you need an extra switch. If you have a spare one or can afford another this is much better, above "will work" but itsn't really recommended, just a design brain fart. Infact if you do have some money to spend, you can do much better than both.

[farisa]

GLBP would be a good option in this scenario if you're not running full Internet route table and iBGP because the best route will always be selected leaving your AS, regardless which router the gateway protocol will bring it to. Note that GLBP will not load balance if it's in front of a firewall because firewalls don't allow the proxy arp that goes on with GLBP.

Your best bet would be to use PfR. It will do exactly what you're typing to accomplish, outbound load balancing across two BGP links.

Sent from Cisco Technical Support iPad App

[jwbensley]

farisa wrote:
GLBP would be a good option in this scenario if you're not running full Internet route table and iBGP because the best route will always be selected leaving your AS, regardless which router the gateway protocol will bring it to. 

In this case both upstream links are from the same provider so presumeably the same routes are advetised down both links so it should work fine.

As you for your comment on;

farisa wrote:
Note that GLBP will not load balance if it's in front of a firewall 

I hope there isn't a firewall in the way, if so, that is a faux pas Shizai Li for not mentioning so.

[farisa]

It will not load balance if iBGP is still in the mix. Only one best path will be used and no load balancing will occur, even with GLBP. iBGP would have to be disabled and then rely on GLBP to do the load balancing between the two routers.

Since the original poster mentioned these are ISP connections, I figured it would be safe to say a firewall is in the mix.

PfR will accomplish the load sharing; no installation of additional switches or BVIs necessary. Works pretty well too.

[jwbensley]

Ah I see, sorry I missunderstood.

PfR, I was reading this as PBR I am not familiar with PfR, I am now reading about it, it's pretty interesting stuff!

Thanks Farisa.

[farisa]

My pleasure.

Yes, it is pretty cool. Solves for the biggest let down of BGP, load balancing!

[michaelca99]

Thank for all responses.

This is a real topology. There is impossible to add any other physical link beween any devices.

Reasons:

1. R1 and R2 in different location, the distance is 60km

2. SW1, SW2 and SW3 are on the top of different mountains.

3. It really difficult (almost impossible) to upgrade the links among SW1, SW2 and Sw3. The cost would be million dollars.

4. Setup the second link will cause 20,000 to 30,000 and easy to do

About balancing:

Currently, all Traffic passes through Link1.We want VLAN 16,15 Traffic passes through Link 2, and VLAN 17,18 and 19 goes through Link 1.  The inbound traffic should come back same as outgoing link.

Firewalll:

There is not firewall in the way.

[farisa]

For incoming traffic, you can prepend the prefixes out the circuit that you less prefer remote traffic to access you throguh. For outbound, you may need to do a combination of local preference based on destination prefix or PBR based on source.