load balanced routing?

[bphelps]

I have two point to point T1 links between two routers that I want to use as one 3 MB circuit.

MLPPP doesn't work because the service provider won't support it.

If I configure two default routes on each router, it is said that traffic will be load balanced. Does anyone know how this happens? Is it per-packet? Would this solution create the illusion of a 3 MB pipe?

[v-nguyen]

I don't know about 2 x default routes and how they "formed" a 3M pipe. Explain that to me more if you don't mind but there are things that you can do such as:

1. Define all class A addresses to be upload/download from one router and leave the rest of class B and C on the other router. You'll have to ask your service provider to do the same thing on their side. This setup you won't have to worry about packet load balancing.

2. Define one router to be the upload router and the other to be the download router. Again you'll have to work together with your ISP on this.

[bphelps]

These are point to point links, they are not links to the internet. To clarify:

Router | | Router

A |-----------------------| B

|-----------------------|

| |

each -------- line represents one T1. Each T1 uses all 24 channels for a combined bandwidth of 3.088 Mbps. What I was thinking is that we could put a different subnet on each T1 line, then place two default route commands on each router, like this:

Router A

ip route 0.0.0.0 0.0.0.0 s1/0

ip route 0.0.0.0 0.0.0.0 s1/1

Router B

ip route 0.0.0.0 0.0.0.0 s1/0

ip route 0.0.0.0 0.0.0.0 s1/1

Cisco says that if you have two default routes on one router, traffic destined for the default route is load balanced between the two. I want to know how that load balancing works, and if it solves the problem using both links fully.

[bphelps]

Okay, that picture didn't turn out right. I'll try to make it look better:

Router A |==============| Router B

Where each --------- represents one T1....

[vcjones]

All you need to do (to load balance traffic from your router to the service provider) is to define two static default routes, one for each T1. This assumes that you are not using NAT and that you can reliably detect at the link level when either link is down. By default, you will get per-destination load balancing unless fast switching is disabled with a "no ip route-cache." Unless both links go to the same router at both ends, you want to use per-destination rather than per-packet. If you want to use per-packet, use CEF and configure it there.

Note that while you can load balance your traffic to the ISP, it is up to the ISP to load balance traffic coming back to you. You have no way to do that without the cooperation of your ISP.

Good luck and have fun!

Vincent C Jones

www.networkingunlimited.com

[wkumari · Accepted solution]

If you have 2 routes to a prefix in the routing table with equal metrics, the router should loadbalance between them. Whether per-packet or per-flow depends (primarily) on your switching (process switching will do per-packet, CEF will prefer per-destination, but can be made to do per-packet). In most cases per-destination load balancing will be the best way to go as it is "lighter" on the router (and in cases where the remote ends terminate on different routers you don't get out-of-order packets). One way to do per-destination if you don't want to run CEF is:

ip route 0.0.0.0 0.0.0.1 S0

ip route 0.0.0.0 0.0.0.0 S1

ip route 0.0.0.0 0.0.0.0 S0 10

As for getting the circuits load-balanced inbound, try get the service provider to do the same thing towards you (but obviously with your address space!)

Most of the above is assuming that your traffic distribution is "normal" (you are not sending all of your traffic to one IP (or only even or odd IP's). One of the nice things with CEF is that it will so a source-destination hash and so (largely) avoid this problem. Here is a nice link:

http://www.cisco.com/warp/public/cc/pd/ifaa/pa/much/prodlit/loadb_an.htm

Warren.

-- Warren.

[raarons]

Load balancing depends on how the routers and interfaces are configured.

If CEF is not possible or switched off, load balancing will be per-destination by default, with fast switching. If fast switching is disabled (i.e. "no ip route-cache"), load balancing will be per-packet, but process-switching 2 T1s will stress out a small router.

With CEF, the default load balancing methos is per-destination again, but per-packet load balancing can be configured with "ip load-sharing per-packet", with no loss in performance. This link may be helpful

http://www.cisco.com/en/US/tech/tk648/tk365/technologies_tech_note09186a0080094806.shtml