LACP and parallel X's

[rockhead]

What I am trying to do is use LACP to run two AFX links in parallel

Code: Select all

 
                     AF5X->AF5X
Netonix12AC /                      \   Netonix 12AC
                   \                      /
                      AF3X->AF3X

 

Has anybody out there done this ? I have tried all the settings I can imagine on the switches with no luck yet, I appear to have been able to achieve failover or looping but not active use of both links ?

[sirhc]

You do realize that using LAGs (Static or LACP) for aggregation is not a wise idea, it can be done but......

Layer 2 devices are only aware of Physical link states so the switches will think there is 1G FD on each pipe.

Meaning the switch will have no idea when 1 link becomes saturated and stop using it in favor of the other link to balance the load.

You can specify some parameters that will be used to balance the load but latency on one link over the other is not available.

LAGs do not combine the 2 pipe/links into a single pipe/link but rather load balance across them.

It will work but there are drawbacks as mentioned above.

Make sure your running v1.3.9 or newer but be aware we are working on some bugs in v1.4.0rc10 such as graphs.

[rockhead]

Actually that clarifies it perfectly, it maybe could be done, but probably should not. I have seen a ton of chatter about this lately, some that even alleged you had done it, LOL.
Looks like I'll go with distinct vlans for now until I am comfortable rolling OSPF onto multiple remote sites

[tma]

Not sure OSPF and equal cost multipath (ECMP) routing will help you better than a LAG (with or without LACP), i.e. OSPF also has no clue about the capacity of each leg. Capacity may be different to start with (as you are mixing AF5X and AF3X) and it may vary under fading conditions or from interference. ECMP won't account for that and will not shift loads, just like a LAG will not.

[sirhc]

Not sure OSPF and equal cost multipath (ECMP) routing will help you better than a LAG (with or without LACP), i.e. OSPF also has no clue about the capacity of each leg. Capacity may be different to start with (as you are mixing AF5X and AF3X) and it may vary under fading conditions or from interference. ECMP won't account for that and will not shift loads, just like a LAG will not.

That is a good point about OSPF, however I believe OSPF v3 is aware but we have not yet implemented it mostly because it relies heavily on IPv6 I think

[cbl]

http://blog.ipspace.net/2007/02/unequal ... aring.html has a couple of hints.. but again, it doesnt account for link fade, etc.