Hello,
It depends on the particular type of technology.
In EtherChannel, frames are sent through the individual Ethernet links depending on the combination of MAC addresses in the frames, optionally even IP addresses and TCP/UDP port numbers. For every frame, the switch feeds these values into a hashing function, and the result of the hashing function points to a particular link through which the frame will be forwarded. As a result, EtherChannel load-balances the traffic across multiple links.
In PPP Multilink, every frame is split into several fragments, one for each link, and the fragments are the sent over the links in parallel. The receiving router will reassemble these fragments back to the original frame and process its payload as usual. As a result, PPP Multilink also achieves load balancing.
Other modes of operation, such as using the second link in a bundle only when the first one is already saturated are possible but uncommon. I recall that on ISDN BRI interfaces, you could configure the dialer to place another call to the ISP if the existing connection was coming near to saturation, and PPP Multilink was used at the backend to add or remove the newly added call to the bundle of calls capable of carrying data. However, I haven't seen this mode of operation with Ethernet.
Best regards,
Peter
