Load balancing across WAN links

[kwillacey]

I have two WAN links that I would like to load balance traffic across using EIGRP but both links come in on a switch card and are separated using VLANS. I know with EIGRP its important to have the delay and bandwidth set to accomplish load balancing.

My question is should I place the bandwidth and delay on the VLAN interface or on the Fast Ethernet interface???

[om.v.verma]

Hi

There is one more option, u can use variance

feture of eigrp. This will resolve ur prob.

Regards

Om

[paolo bevilacqua]

Excuse me, have you ever used the variance supposed functionality? Be aware that beside requiring special configuration, it never worked in practice, and has no know implementation in customers network.

[om.v.verma]

ok ok i got ur question

U have to configure all the parameter on the

fastethernet interface.

Regards

Om

[paolo bevilacqua]

Hi,

please don't change the EIGRP default values before determining a need for that. In a layer-3 switch, all the IP level configuration is made at the VLAN interface.

[kwillacey]

Well it is actually a router with a switch card so I am to assume I should place the statments on the VLAN interface then.

[paolo bevilacqua]

Yes. Under the physical interface, you only configure to which VLAN it pertains, and that's it.

Hope this helps, please rate post if it does!

[kwillacey]

So to clarify my config should reflect below:

int vlan 2

ip address 1.1.1.1 255.255.255.0

bandwidth 100

delay 100

int fa0/0/0

switchport access vlan 2

the bandwidth and delay statements should not go on the int fa0/0/0

[paolo bevilacqua]

Hi,

As I said above you need not to configure bandwidth and delay and your eigrp will work fine. The parameters influence the metric compute, thing that you don't necessarily need to do.

[Edison Ortiz]

Kelvin,

I agree with Paolo regarding the EIGRP configuration. Before changing the metrics, can you post a show ip route [network.address] from a route being received via the 2 VLANs ?

This way we can recommend if you need the bandwidth and delay statements in the configuration. Playing with these metric without checking your initial configuration, can create future problems.

And to answer your question, all metric values should be placed on the interface running the protocol. The interface running the protocol must hold an IP address. Per your previous post, a FastEthernet interface is only layer 2 - the SVI (Switch Virtual Interface) is the one holding the L3 information. However, please post the show ip route before proceeding.

Thanks

[jay77jay77]

As said, i will suggest not to change any metrics, as both the interfaces are similar and metrics will be the same and this will let eigrp to loadbalance the traffic.

Also as you told, the interfaces are on the same router, how does it terminate at the remote end... on the same router??

If so, you may test out per-packet loadbalancing for complete 50/50 load balancing.

If not eigrp will do load balancing per destination, whic lead to uneven loadbalancing..

Note: per packet loadbalancing consumes more processor cycles...

So the selection configuration of perpacket or perdestination depends on the traffic pattern across the two sites.

Cheers!

[paolo bevilacqua]

hello jay,

please note, per se, eigrp or other routing protocols do not decide the type of load balancing done, even if some documentation may lead you to think so. Layer3 switch have their own balancing algos that are basically similar to CEF in routers. There are no process cycles in layer3 switches as the routing decision is made in hardware.

I also suggest to not enable per packet load-balacing as it may lead to out-of-sequence arrival.

[bgibson]

This is the biggest issue with per packet load balancing ESPECIALLY if you are doing any kind of VoIP or multimedia over IP for that matter.

Personally I recommend you look for other ways to load balance both circuits.

[jay77jay77]

hi gibson,paolo,

do u see out of sequence issue, even on parallel links terminating on the same physical routers at both ends?

Cheers!

[bgibson]

It's almost unavoidable. Even in that scenario the packets are entering the queues via different interfaces and treated differently. Throw any kind of load onto one of the circuits or both and your voice will be toast.

Brian Gibson