IP over Dual OC3 Circuits with Load Balancing?

Unanswered Question
Paolo Bevilacqua Mon, 02/09/2009 - 15:25

Either an high-spec 7200 (NPE-Gx), 7300, or 7600. In all cases you will configure MLPPP.

Watch out anyway, because that is not a common config (high speed MLPPP) and you may run into issues.

marikakis Tue, 02/10/2009 - 01:39


We had to do this for one of our POPs (once upon a time). The 2 OC3 links were between 2 routers (one 7500 router at central and one 7200 router at remote site). We did not use MLPPP. We were running OSPF (equal bandwidth for both links configured at the interfaces at both endpoints). We left CEF to the default per-source/destination load-balancing behavior. We monitored the behavior and it turned out to satisfy our needs. This method of load-balancing is based on routing and actual switching method used by routers.

If you try something like this, the outcome depends on the traffic patterns in your network. Load-balancing is not an easy thing to accomplish and maybe successful for sometime until it fails. In any case, I would try the easy things first and go for more drastic measures if it is proven that there is no easy way out.

Kind Regards,


Giuseppe Larosa Tue, 02/10/2009 - 03:02

Hello Maria,

I agree I don't think using ML PPP with two OC-3 speed links can work well.

Hope to help


marikakis Tue, 02/10/2009 - 04:49


It seems we all have our doubts about MPPPP. Just one more memory (I feel like an old lady today :-). We had 2 ML PPP bundles with 4 E1's each between 2 7200's (NPE-300 or 400, can't remember that). When we replaced the bundles with a single E3 circuit (using PA-E3), the router CPUs took a nice deep breath.

Kind Regards,


Paolo Bevilacqua Tue, 02/10/2009 - 04:53

The concerns about high speed MLPPP are justified, mainly by experience.

On the other hand, people insistently asks for single-session circuit aggregation, provided the platform is performant enough, and the code is correctly written, there is no reason it should not work.

Joseph W. Doherty Tue, 02/10/2009 - 06:44

In theory, if the device supports MLPPP on the interfaces, no reason it shouldn't work, the concern, of course, is additional processing load. This is of special concern for the platforms that provide "high performance" by doing much of their normal packet forwarding via hardware.

I haven't tested it, but suspect the additional load of MLPPP might depend much on whether you have it fragment packets across the links.

I agree with the other posters who recommend routing across both paths. You could even use CEF in packet-by-packet mode, although at the risk flow sequencing issues.

Depending on what your topology is, another possible solution might be usage of OER/PfR and have it dynamically balance flows across the dual links. (This is still a routed solution, but can avoid both the CEF packet-by-packet issue and asymetric load balancing that can arise with normal CEF.)


This Discussion