02-11-2025 12:00 AM
Hi Everyone
I'm using ASR920 and ASR900 platforms for MPLS and metro environment and I'm facing a weird behavior on their etherchannels.
for example i have an ASR920 PE with 2x10G bundled links to a C4500x.
in the ASR920 i have set the load-balancing method to source-mac and configured 2 service-instances on the port-channel with different mac-addresses on the BD interface.
Right now there is about 1Gbps of MPLS L3VPN traffic on the router.
The traffic is balancing unevenly on the BDIs (BDI100,101) using CEF but when it comes to the etherchannel, it is balancing evenly(Te0/0/12-13).
show int su | i \*
*: interface is up
* Te0/0/12 0 0 0 0 311583000 45534 350212000 68489 0
* Te0/0/13 0 0 0 0 585939000 113863 550976000 90881 0
* GigabitEthernet0 0 0 0 0 0 0 0 0 0
* BDI100 0 0 0 0 23894000 11140 647694000 81645 0
* BDI101 0 0 0 0 136421000 49584 4123000 736 0
* Loopback0 0 0 0 0 0 0 0 0 0
* Port-channel6 0 0 0 0 897522000 159397 901188000 159370 0
i have configured source-mac balancing so these are the expected behaviors:
1. Traffic of both BDIs go through 1 of the 10G links.
2. One BD goes through Te0/0/12 and the other through Te0/0/13
right now it is not doing any of those and actually balancing the traffic better than i expected.
but i wonder why is this happening?
i can see the exact same behavior on ASR903 with 4x10G bundled interfaces and configured with source-mac balancing.
It is balancing perfectly even though there is only 2 MAC in the bridge-domain and the router is switching between them.(It is normal IP. traffic should go through one link)
I am very glad that it is balancing so good but why isn't the router balancing as i have configured it.
show etherchannel load-balancing
EtherChannel Load-Balancing Method:
Global LB Method: flow-based
LB Algo type: Source MAC
Port-Channel: LB Method
Port-channel6 : flow-based (Source MAC)
Thanks
Amir
02-11-2025 07:09 AM
It appears to me that you have traffic on Po16 (and its links Te0/0/12 & Te0/0/13) that is not accounted for by the sum of the traffic from BDI100 + BDI101. That is, the bps and pps from Te0/0/12 & Te0/0/13 sum up to the bps and pps for Po16, but these totals are considerably greater than the totals of BDI100 + BDI101. Where might this extra traffic be coming from?
That aside, there is absolutely no guarantee that flows from two different source MACs will hash to two different links in a LAG bundle. With a hash calculation resulting in a random distribution (the ideal result), there is a 50% chance that all flows will hash to the same link in a 2-link bundle when using Source MAC with two sources.
Discover and save your favorite ideas. Come back to expert answers, step-by-step guides, recent topics, and more.
New here? Get started with these tips. How to use Community New member guide