1. I think some of the following statements from the referenced blog post are wrong:

    > In the case of unequal-cost multipathing (also called “unequal-cost load balancing” by some technologies, such as EIGRP), you will have two paths. > Path 1 might have a lower metric, so internally it’ll be given hashes 0 and 1. > Path 2 might have a higher metric, so internally it’ll be given hashes 2, 3, 4, 5, 6, and 7.

    Paths with lower metric are usually more preferred. So really it should be the opposite, shouldn't it?

    Assuming a metric (cost) ratio of 1:3, path 1 will be associated with hashes 0, 1, 2, 3, 4, 5 whereas path 2 will be linked to hashes 6 and 7.

    1. Of course you're right. Lower end-to-end cost should get more hash entries... but then there's BGP DMZ BW where higher bandwidth gets you more hash entries. Same with port channels.

      Will try to contact the author of the blog post ;)

    2. Thanks for engaging here. It makes sense that paths with higher bandwidth are more preferred (more hashes attached). Do you mean one could configure bandwidth command on individual interfaces belonging to a port channel? I thought that interfaces belonging to a port channel must be of the same speed and duplex (at least on Cisco devices). I'm aware that speed is unlike bandwidth. I'm not aware of unequal load balancing on port channels.

Add comment