7206vxr Load-Sharing

Unanswered Question
Nov 5th, 2007

I currently have 2 offices with 16 Pt-Pt T1's. These are split in to 2 Multilinks, each with 8 T1's in each. We split this so that replication traffic is on 1 multilink and general data is on another. I am seeing that we are not fully utilizing our BW and would like to consider load-sharing between the two. Is this possible in our setup?

-Both routers are using 12.4

-EIGRP is used between sites

-ip CEF is enabled

Would I apply 'ip load-sharing per-packet' in the Multilink Interface or remove the multilink and apply the command in the Serial Interface? or neither?

I have this problem too.
0 votes
  • 1
  • 2
  • 3
  • 4
  • 5
Overall Rating: 5 (1 ratings)
Richard Burts Mon, 11/05/2007 - 09:14


You would not want to remove the multilink. If you did so then the limitation on max-links in EIGRP would prevent you using all of the 16 links.

You should be able to load share over both multilinks if that is what you want to do. We could advise better how to do this if we knew what you had done to separate the replication traffic from the general data traffic. Without knowing how the traffic is separated our advise might not work effectively.

I would suggest that you think about the implications of not separating the traffic. Someone made the decision to provision the T1s and to separate the traffic. I would assume that this was to prevent the replication traffic from impacting general data traffic. If you remove the separation and begin to load share what is the potential impact on your general data traffic?



Snydersh1_2 Mon, 11/05/2007 - 11:22


The seperation of traffic was done prior to me being here. Since then though, I installed the VXR and applied generic static routes for the replication traffic to use the 2nd multilink. The route is based on destination. If we began to load-share, could we apply a QoS policy to help throttle the replication if need be? Also, the replication usually occurs after production hours so impact to general data during business hours would be minimal.


paul.matthews Mon, 11/05/2007 - 11:55

QoS would indeed be of use in preventing replication traffic from swamping general traffic.

As it stands, you have the potential of two layers of multilink - PPP within each multilink, and eigrp across the two.

What is the actual topology - do you use a pair of VXRs at the central site and a pair of routers at the remote site. or is it a single router at remote sites and dual routers at central? even a single router at central?

Depending on the details, you can look at a number of options to load balance across both.

If it is a single router at each end, consider just bundling both links into one MLPPP bundle. Use QoS on the single logical interface and it is tidy.

If multiple routers at either end, then the load balancing via EIGRP is the best solution (you may prefer it anyway to one big bundle) though if multiple routers you will need to consider HSRP/GLBP policy, and configure variance on eigrp. There is the risk though that using varince may lead to the possibility of the odd packet being passed between the two local routers if that is the way you ae doing it.

From the sounds of things, the chronological separation should be OK, and you could perhaps just let EIGRP do its think, and use PBR to force the replication traffic to use a single link. You could even look at time based access lists to allow it to use as much bandwidth as it wants between certain hours, and a single link during business hours.


Richard Burts Mon, 11/05/2007 - 12:02


You probably could configure QoS to throttle replication data. Would the impact on replication be acceptable?

Do I understand correctly that EIGRP is running over 1 of the multilinks (providing routing for the general data traffic) and that EIGRP is not running on the second multilink. And that a static route is used to send replication data on the second multilink? If so then beginning to load share is probably as easy as putting the second multilink into EIGRP. If EIGRP is running on both multilinks then general data traffic should flow on both multilinks.



Snydersh1_2 Mon, 11/05/2007 - 12:14


We have a single VXR at each location w/ 2 multilinks. Multilink 1 has a defined bandwidth of 12000 and Multilink 2 has a defined bandwidth of 11000. I did this so that both Multilinks are in EIGRP and if one of the Multilinks failed, EIGRP would route over the Up-Up Multilink. I have static routes pointing my replication data over Multilink 2 w/ a floating static route for redundancy.


Are you saying that I can bundle Multilink 1 and Multilink 2 to a single MLPPP and use QoS?


paul.matthews Mon, 11/05/2007 - 12:23

Yes, I am saying that it is a possibility - I am not necessarily saying it is the right thing to do though!

From what you have just said, you effectiveky have a single router with two links between them at L3. the "simple" way is to set both links back to the same bandwidth (a little hint, if you want to influence eigrp use delay, not bandwidth) and it will load balance, then use QoS to contain the replication.

Snydersh1_2 Mon, 11/05/2007 - 12:29

Thanks for your help Paul. I'll try to set this up and hopefully I can let you know how it goes.



This Discussion