Hello Po
When we talk about load balancing between physical connections of etherchannels, we must use some criteria to determine which physical link a particular frame will go over. Take a look at the following diagram:
Devices in Network A will be communicating with devices in Network B. Each of the frames traversing the link will have a source and destination MAC, and at Layer 3, will have a source and destination IP address.
Remember first of all that load balancing must be employed on a per direction basis. This means that for the above topology, you must configure the load balancing that will be employed for each direction. For traffic from SW1 to SW2, you configure the load balancing on SW1. For traffic from SW2 to SW1, you configure the load balancing on SW2.
Now let’s say you have 60 hosts on Network A, and you have a single host, say a default gateway, on Network B.
Let’s take a look at traffic going from Network A to Network B. If load balancing by source MAC address is configured, then the traffic going from all the hosts on Network A over the etherchannel, will be load balanced well. Specifically, traffic from 10 hosts will use each single physical connection. (60 source MAC addresses, 6 physical links). That’s good right? But what if you used the destination MAC as the criteria? There’s only one destination MAC on Network B, which is the default gateway. So whenever SW1 sees that destination MAC, it will always send the traffic over a single physical link.
So the source and destination MAC and IP addresses used for the algorithm are those found within the frame/packet for the particular direction of travel of that packet.
You might find these posts helpful as well:
I hope this has been helpful!
Laz