Ive been seeing something weird with how the traffic is sent back and forth through the mikrotik and cisco. it looks like on the cisco that all the traffic is going through just one of the interfaces and when i look at the mikrotik it looks like its actively using both. Im going to port what my config looks like and how the interfaces are behaving, I was hoping to get some insights on how this works and how to properly set this up. I would like to make sure that all connections use the same link but still as new connections travel through the bonded (portchanneled) interface that it is spread out along both interfaces to take advantage of both interfaces.
Any insight would be greatly appreciated. 1st image shows the interfaces.. the first two look like they are somehow splitting the traffic but to one interface? i dont see how that can happen.
THANKS AGAIN!!
The bond member through which data is sent is determined purely by the sending end - so Cisco for Cisco->Mikrotik packets and Mikrotik for Mikrotik->Cisco packets.
Using Source MAC address for EtherChannel Load Balancing on the Cisco appears to be the worst choice for your use case, and layer 2 on the Mikrotik is not great either.
Part of what has been happened is that two different connections running iPerf at the same time exhaust one link. Two iPerf servers on another switch: and two iPerf clients on the end with of a Cisco and one computer gets the full connection and the other just 2 megs. The other connection doesn’t begin on the different links that’s what I’m trying to understand..
Thanks for being so gracious with your response btw..
Using Source MAC address for EtherChannel Load Balancing on the Cisco appears to be the worst choice for your use case, and layer 2 on the Mikrotik is not great either.
Using Layer2 is not wrong at all…
Most effective would be Layer2+Layer3…
I was thinking more of source+destination MAC being a better choice than source MAC only, which is the Cisco default, for layer 2. The Cisco may only support MAC (src, dst, src+dst) OR IP (src, dst, src+dst) methods, not any combined MAC+IP / layer2+layer3 ones which are often better.
I’d go for L3 then … any traffic towards default gateway will always have the same dst MAC address (the one of gateway) while it’ll have lots of different dst IP addresses. Same for return traffic (and src MAC or IP addresses).
For LAN-LAN traffic it’s mostly the same, mostly computers have single IP address and single MAC address.
To maximize multiple physical links of a bond it’s necessary to use Tx strategy which gives the most different possibilities. L4 (src and dst ports) adds some entropy, sadly most bonding algorithms don’t support using L4 in Tx strategy.
